fse - pypi Package Compare versions

Comparing version 0.1.15 to 0.1.17

fse.egg-info/PKG-INFO

 Metadata-Version: 1.0
 Name: fse
-Version: 0.1.15
+Version: 0.1.17
 Summary: Fast Sentence Embeddings for Gensim
-Home-page: https://github.com/oborchers/Fast_Sentence_Embeddings
+Home-page: UNKNOWN
 Author: Oliver Borchers
-Author-email: [e-mail]
-License: GPL-3.0
+Author-email: [e-mail]
+License: UNKNOWN
 Description: UNKNOWN
 Platform: UNKNOWN

fse.egg-info/requires.txt

@@ -5,4 +5,4 @@ numpy>=1.11.3
 scikit-learn>=0.19.1
-gensim>=3.8.0
+gensim<4
 wordfreq>=2.2.1
 psutil

fse.egg-info/SOURCES.txt

@@ -1,9 +0,4 @@
-.gitattributes
-.gitignore
-LICENSE
 MANIFEST.in
 README.md
 setup.py
-./fse/models/average_inner.c
-./fse/models/average_inner.pyx
 fse/__init__.py
@@ -36,3 +31,2 @@ fse/inputs.py
 fse/test/test_utils.py
-fse/test/test_data/.DS_Store
 fse/test/test_data/test_sentences.txt

fse/__init__.py

@@ -0,13 +1,16 @@
+import logging
+
 from fse import models
 
-from .inputs import BaseIndexedList
-from .inputs import IndexedList
-from .inputs import CIndexedList
-from .inputs import SplitIndexedList
-from .inputs import SplitCIndexedList
-from .inputs import CSplitIndexedList
-from .inputs import CSplitCIndexedList
-from .inputs import IndexedLineDocument
+from .inputs import (
+    BaseIndexedList,
+    CIndexedList,
+    CSplitCIndexedList,
+    CSplitIndexedList,
+    IndexedLineDocument,
+    IndexedList,
+    SplitCIndexedList,
+    SplitIndexedList,
+)
 
-import logging
 
@@ -18,4 +21,5 @@ class NullHandler(logging.Handler):
 
-logger = logging.getLogger('fse')
+
+logger = logging.getLogger("fse")
 if len(logger.handlers) == 0:  # To ensure reload() doesn't add another one
     logger.addHandler(NullHandler())

fse/inputs.py

 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
 
-# Author: Oliver Borchers <[e-mail]>
-# Copyright (C) 2019 Oliver Borchers
+# Author: Oliver Borchers
+# Copyright (C) Oliver Borchers Oliver Borchers
 
-from typing import NamedTuple, List, MutableSequence
+from pathlib import Path
+from typing import List, MutableSequence, Union
 
 from gensim.utils import any2unicode
+from numpy import concatenate, ndarray
 from smart_open import open
 
-from pathlib import Path
 
-from numpy import ndarray, concatenate
-
 class BaseIndexedList(MutableSequence):
+    def __init__(self, *args: List[Union[list, set, ndarray]]):
+        """Base object to be used for feeding in-memory stored lists of sentences to the
+        training routine.
 
-    def __init__(self, *args:[list, set, ndarray]):
-        """ Base object to be used for feeding in-memory stored lists of sentences to
-        the training routine.
-
         Parameters
@@ -26,7 +24,6 @@ ----------
             Arguments to be merged into a single contianer. Can be single or multiple list/set/ndarray objects.
-
         """
 
         self.items = list()
-        
+
         if len(args) == 1:
@@ -41,4 +38,4 @@ self._check_list_type(args[0])
 
-    def _check_list_type(self, obj:object):
-        """ Checks input validity """
+    def _check_list_type(self, obj: object):
+        """Checks input validity."""
         if isinstance(obj, (list, set, ndarray)):
@@ -49,4 +46,4 @@ return 1
 
-    def _check_str_type(self, obj:object):
-        """ Checks input validity """
+    def _check_str_type(self, obj: object):
+        """Checks input validity."""
         if isinstance(obj, str):
@@ -58,4 +55,4 @@ return 1
     def __len__(self):
-        """ List length 
-        
+        """List length.
+
         Returns
@@ -75,9 +72,8 @@ -------
            Human readable representation of the object's state (words and tags).
-
         """
         return str(self.items)
 
-    def __getitem__(self, i:int) -> tuple:
-        """  Getitem method
-        
+    def __getitem__(self, i: int) -> tuple:
+        """Getitem method.
+
         Returns
@@ -90,23 +86,23 @@ -------
 
-    def __delitem__(self, i:int):
-        """ Delete an item """
+    def __delitem__(self, i: int):
+        """Delete an item."""
         del self.items[i]
-        
-    def __setitem__(self, i:int, item:str):
-        """ Sets an item """
+
+    def __setitem__(self, i: int, item: str):
+        """Sets an item."""
         self._check_str_type(item)
         self.items[i] = item
 
-    def insert(self, i:int, item:str):
-        """ Inserts an item at a position """
+    def insert(self, i: int, item: str):
+        """Inserts an item at a position."""
         self._check_str_type(item)
         self.items.insert(i, item)
 
-    def append(self, item:str):
-        """ Appends item at last position"""
+    def append(self, item: str):
+        """Appends item at last position."""
         self._check_str_type(item)
         self.insert(len(self.items), item)
-    
-    def extend(self, arg:[list, set, ndarray]):
-        """ Extens list """
+
+    def extend(self, arg: [list, set, ndarray]):
+        """Extens list."""
         self._check_list_type(arg)
@@ -119,8 +115,8 @@
 
+
 class IndexedList(BaseIndexedList):
+    def __init__(self, *args: [list, set, ndarray]):
+        """Quasi-list to be used for feeding in-memory stored lists of sentences to the
+        training routine.
 
-    def __init__(self, *args:[list, set, ndarray]):
-        """ Quasi-list to be used for feeding in-memory stored lists of sentences to
-        the training routine.
-
         Parameters
@@ -130,9 +126,8 @@ ----------
             Arguments to be merged into a single contianer. Can be single or multiple list/set objects.
-
         """
         super(IndexedList, self).__init__(*args)
 
-    def __getitem__(self, i:int) -> tuple:
-        """  Getitem method
-        
+    def __getitem__(self, i: int) -> tuple:
+        """Getitem method.
+
         Returns
@@ -145,8 +140,8 @@ -------
 
+
 class CIndexedList(BaseIndexedList):
+    def __init__(self, *args: [list, set, ndarray], custom_index: [list, ndarray]):
+        """Quasi-list with custom indices to be used for feeding in-memory stored lists
+        of sentences to the training routine.
 
-    def __init__(self, *args:[list, set, ndarray], custom_index:[list, ndarray]):
-        """ Quasi-list with custom indices to be used for feeding in-memory stored lists of sentences to
-        the training routine.
-
         Parameters
@@ -158,3 +153,2 @@ ----------
             Custom index to support many to one mappings.
-
         """
@@ -166,7 +160,9 @@ self.custom_index = custom_index
         if len(self.items) != len(self.custom_index):
-            raise RuntimeError(f"Size of custom_index {len(custom_index)} does not match items {len(self.items)}")
+            raise RuntimeError(
+                f"Size of custom_index {len(custom_index)} does not match items {len(self.items)}"
+            )
 
-    def __getitem__(self, i:int) -> tuple:
-        """  Getitem method
-        
+    def __getitem__(self, i: int) -> tuple:
+        """Getitem method.
+
         Returns
@@ -179,23 +175,23 @@ -------
 
-    def __delitem__(self, i:int):
+    def __delitem__(self, i: int):
         raise NotImplementedError("Method currently not supported")
-        
-    def __setitem__(self, i:int, item:str):
+
+    def __setitem__(self, i: int, item: str):
         raise NotImplementedError("Method currently not supported")
 
-    def insert(self, i:int, item:str):
+    def insert(self, i: int, item: str):
         raise NotImplementedError("Method currently not supported")
 
-    def append(self, item:str):
+    def append(self, item: str):
         raise NotImplementedError("Method currently not supported")
-    
-    def extend(self, arg:[list, set, ndarray]):
+
+    def extend(self, arg: [list, set, ndarray]):
         raise NotImplementedError("Method currently not supported")
 
+
 class SplitIndexedList(BaseIndexedList):
+    def __init__(self, *args: [list, set, ndarray]):
+        """Quasi-list with string splitting to be used for feeding in-memory stored
+        lists of sentences to the training routine.
 
-    def __init__(self, *args:[list, set, ndarray]):
-        """ Quasi-list with string splitting to be used for feeding in-memory stored lists of sentences to
-        the training routine.
-
         Parameters
@@ -205,9 +201,8 @@ ----------
             Arguments to be merged into a single contianer. Can be single or multiple list/set objects.
-
         """
         super(SplitIndexedList, self).__init__(*args)
 
-    def __getitem__(self, i:int) -> tuple:
-        """  Getitem method
-        
+    def __getitem__(self, i: int) -> tuple:
+        """Getitem method.
+
         Returns
@@ -220,8 +215,8 @@ -------
 
+
 class SplitCIndexedList(BaseIndexedList):
+    def __init__(self, *args: [list, set, ndarray], custom_index: [list, ndarray]):
+        """Quasi-list with custom indices and string splitting to be used for feeding
+        in-memory stored lists of sentences to the training routine.
 
-    def __init__(self, *args:[list, set, ndarray], custom_index:[list, ndarray]):
-        """ Quasi-list with custom indices and string splitting to be used for feeding in-memory stored lists of sentences to
-        the training routine.
-
         Parameters
@@ -233,3 +228,2 @@ ----------
             Custom index to support many to one mappings.
-
         """
@@ -241,8 +235,9 @@ self.custom_index = custom_index
         if len(self.items) != len(self.custom_index):
-            raise RuntimeError(f"Size of custom_index {len(custom_index)} does not match items {len(self.items)}")
+            raise RuntimeError(
+                f"Size of custom_index {len(custom_index)} does not match items {len(self.items)}"
+            )
 
+    def __getitem__(self, i: int) -> tuple:
+        """Getitem method.
 
-    def __getitem__(self, i:int) -> tuple:
-        """  Getitem method
-        
         Returns
@@ -254,24 +249,24 @@ -------
         return (self.items.__getitem__(i).split(), self.custom_index[i])
-    
-    def __delitem__(self, i:int):
+
+    def __delitem__(self, i: int):
         raise NotImplementedError("Method currently not supported")
-        
-    def __setitem__(self, i:int, item:str):
+
+    def __setitem__(self, i: int, item: str):
         raise NotImplementedError("Method currently not supported")
 
-    def insert(self, i:int, item:str):
+    def insert(self, i: int, item: str):
         raise NotImplementedError("Method currently not supported")
 
-    def append(self, item:str):
+    def append(self, item: str):
         raise NotImplementedError("Method currently not supported")
-    
-    def extend(self, arg:[list, set, ndarray]):
+
+    def extend(self, arg: [list, set, ndarray]):
         raise NotImplementedError("Method currently not supported")
 
+
 class CSplitIndexedList(BaseIndexedList):
+    def __init__(self, *args: [list, set, ndarray], custom_split: callable):
+        """Quasi-list with custom string splitting to be used for feeding in-memory
+        stored lists of sentences to the training routine.
 
-    def __init__(self, *args:[list, set, ndarray], custom_split:callable):
-        """ Quasi-list with custom string splitting to be used for feeding in-memory stored lists of sentences to
-        the training routine.
-
         Parameters
@@ -283,3 +278,2 @@ ----------
             Split function to be used to convert strings into list of str.
-
         """
@@ -289,5 +283,5 @@ self.custom_split = custom_split
 
-    def __getitem__(self, i:int) -> tuple:
-        """  Getitem method
-        
+    def __getitem__(self, i: int) -> tuple:
+        """Getitem method.
+
         Returns
@@ -300,8 +294,13 @@ -------
 
+
 class CSplitCIndexedList(BaseIndexedList):
+    def __init__(
+        self,
+        *args: [list, set, ndarray],
+        custom_split: callable,
+        custom_index: [list, ndarray],
+    ):
+        """Quasi-list with custom indices and ustom string splitting to be used for
+        feeding in-memory stored lists of sentences to the training routine.
 
-    def __init__(self, *args:[list, set, ndarray], custom_split:callable, custom_index:[list, ndarray]):
-        """ Quasi-list with custom indices and ustom string splitting to be used for feeding in-memory stored lists of sentences to
-        the training routine.
-
         Parameters
@@ -315,15 +314,16 @@ ----------
             Custom index to support many to one mappings.
-
         """
         self.custom_split = custom_split
         self.custom_index = custom_index
-        
+
         super(CSplitCIndexedList, self).__init__(*args)
 
         if len(self.items) != len(self.custom_index):
-            raise RuntimeError(f"Size of custom_index {len(custom_index)} does not match items {len(self.items)}")
+            raise RuntimeError(
+                f"Size of custom_index {len(custom_index)} does not match items {len(self.items)}"
+            )
 
-    def __getitem__(self, i:int) -> tuple:
-        """  Getitem method
-        
+    def __getitem__(self, i: int) -> tuple:
+        """Getitem method.
+
         Returns
@@ -336,21 +336,21 @@ -------
 
-    def __delitem__(self, i:int):
+    def __delitem__(self, i: int):
         raise NotImplementedError("Method currently not supported")
-        
-    def __setitem__(self, i:int, item:str):
+
+    def __setitem__(self, i: int, item: str):
         raise NotImplementedError("Method currently not supported")
 
-    def insert(self, i:int, item:str):
+    def insert(self, i: int, item: str):
         raise NotImplementedError("Method currently not supported")
 
-    def append(self, item:str):
+    def append(self, item: str):
         raise NotImplementedError("Method currently not supported")
-    
-    def extend(self, arg:[list, set, ndarray]):
+
+    def extend(self, arg: [list, set, ndarray]):
         raise NotImplementedError("Method currently not supported")
 
+
 class IndexedLineDocument(object):
-
     def __init__(self, path, get_able=True):
-        """ Iterate over a file that contains sentences: one line = tuple([str], int).
+        """Iterate over a file that contains sentences: one line = tuple([str], int).
 
@@ -376,5 +376,5 @@ Words are expected to be already preprocessed and separated by whitespace. Sentence tags are constructed
             self._build_offsets()
-    
+
     def _build_offsets(self):
-        """ Builds an offset table to index the file """
+        """Builds an offset table to index the file."""
         with open(self.path, "rb") as f:
@@ -385,7 +385,7 @@ offset = f.tell()
                 offset += len(line)
-    
+
     def __getitem__(self, i):
-        """ Returns the line indexed by i. Primarily used for 
+        """Returns the line indexed by i. Primarily used for
         :meth:`~fse.models.sentencevectors.SentenceVectors.most_similar`
-        
+
         Parameters
@@ -403,3 +403,5 @@ ----------
         if not self.get_able:
-            raise RuntimeError("To index the lines, you must contruct with get_able=True")
+            raise RuntimeError(
+                "To index the lines, you must contruct with get_able=True"
+            )
 
@@ -419,6 +421,5 @@ with open(self.path, "rb") as f:
             Tuple of list of string and index
-
         """
         with open(self.path, "rb") as f:
             for i, line in enumerate(f):
-                yield (any2unicode(line).split(), i)
+                yield (any2unicode(line).split(), i)

fse/models/__init__.py

 from .average import Average
-from .sif import SIF 
+from .sif import SIF
 from .usif import uSIF
-from .sentencevectors import SentenceVectors
+from .sentencevectors import SentenceVectors

fse/models/average_inner.pxd

@@ -7,4 +7,4 @@ # cython: boundscheck=False
 
-# Author: Oliver Borchers <[e-mail]>
-# Copyright (C) 2019 Oliver Borchers
+# Author: Oliver Borchers 
+# Copyright (C) Oliver Borchers Oliver Borchers
 
@@ -11,0 +11,0 @@ cimport numpy as np

fse/models/average_inner.pyx

@@ -8,4 +8,4 @@ #!/usr/bin/env cython
 
-# Author: Oliver Borchers <[e-mail]>
-# Copyright (C) 2019 Oliver Borchers
+# Author: Oliver Borchers 
+# Copyright (C) Oliver Borchers Oliver Borchers
 
@@ -12,0 +12,0 @@ """Optimized cython functions for computing sentence embeddings"""

fse/models/average.py

 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
 
-# Author: Oliver Borchers <[e-mail]>
-# Copyright (C) 2019 Oliver Borchers
+# Author: Oliver Borchers
+# Copyright (C) Oliver Borchers Oliver Borchers
 
@@ -33,3 +33,3 @@ """This module implements the base class to compute average representations for sentences, using highly optimized C routines,
 
-from __future__ import division 
+from __future__ import division
 
@@ -41,4 +41,10 @@ from fse.models.base_s2v import BaseSentence2VecModel
 
-from numpy import ndarray, float32 as REAL, sum as np_sum, multiply as np_mult,\
-    zeros, max as np_max
+from numpy import (
+    ndarray,
+    float32 as REAL,
+    sum as np_sum,
+    multiply as np_mult,
+    zeros,
+    max as np_max,
+)
 
@@ -51,3 +57,9 @@ from typing import List
 
-def train_average_np(model:BaseSentence2VecModel, indexed_sentences:List[tuple], target:ndarray, memory:ndarray) -> [int,int]:
+
+def train_average_np(
+    model: BaseSentence2VecModel,
+    indexed_sentences: List[tuple],
+    target: ndarray,
+    memory: ndarray,
+) -> [int, int]:
     """Training on a sequence of sentences and update the target ndarray.
@@ -77,3 +89,3 @@
     int, int
-        Number of effective sentences (non-zero) and effective words in the vocabulary used 
+        Number of effective sentences (non-zero) and effective words in the vocabulary used
         during training the sentence embedding.
@@ -93,3 +105,2 @@
     mem = memory[0]
-    subwords_idx = memory[1]
 
@@ -114,6 +125,6 @@ if is_ft:
         for obj in indexed_sentences:
-            mem.fill(0.)
+            mem.fill(0.0)
             sent = obj[0]
             sent_adr = obj[1]
-            
+
             word_indices = [vocab[word].index for word in sent if word in vocab]
@@ -125,11 +136,14 @@ eff_sentences += 1
 
-            mem += np_sum(np_mult(w_vectors[word_indices],w_weights[word_indices][:,None]) , axis=0)
-            mem *= 1/len(word_indices)
+            mem += np_sum(
+                np_mult(w_vectors[word_indices], w_weights[word_indices][:, None]),
+                axis=0,
+            )
+            mem *= 1 / len(word_indices)
             s_vectors[sent_adr] = mem.astype(REAL)
     else:
         for obj in indexed_sentences:
-            mem.fill(0.)
+            mem.fill(0.0)
             sent = obj[0]
             sent_adr = obj[1]
-            
+
             if not len(sent):
@@ -140,3 +154,3 @@ continue
             eff_sentences += 1
-            eff_words += len(sent) # Counts everything in the sentence
+            eff_words += len(sent)  # Counts everything in the sentence
 
@@ -148,6 +162,10 @@ for word in sent:
                 else:
-                    ngram_hashes = ft_ngram_hashes(word, min_n, max_n, bucket, True)[:max_ngrams]
+                    ngram_hashes = ft_ngram_hashes(word, min_n, max_n, bucket, True)[
+                        :max_ngrams
+                    ]
                     if len(ngram_hashes) == 0:
                         continue
-                    mem += oov_weight * (np_sum(ngram_vectors[ngram_hashes], axis=0) / len(ngram_hashes))
+                    mem += oov_weight * (
+                        np_sum(ngram_vectors[ngram_hashes], axis=0) / len(ngram_hashes)
+                    )
                 # Implicit addition of zero if oov does not contain any ngrams
@@ -158,5 +176,11 @@ s_vectors[sent_adr] = mem / len(sent)
 
+
 try:
     from fse.models.average_inner import train_average_cy
-    from fse.models.average_inner import FAST_VERSION, MAX_WORDS_IN_BATCH, MAX_NGRAMS_IN_BATCH
+    from fse.models.average_inner import (
+        FAST_VERSION,
+        MAX_WORDS_IN_BATCH,
+        MAX_NGRAMS_IN_BATCH,
+    )
+
     train_average = train_average_cy
@@ -169,4 +193,5 @@ except ImportError:
 
+
 class Average(BaseSentence2VecModel):
-    """ Train, use and evaluate averaged sentence vectors.
+    """Train, use and evaluate averaged sentence vectors.
 
@@ -183,15 +208,23 @@ The model can be stored/loaded via its :meth:`~fse.models.average.Average.save` and
         directly to query those embeddings in various ways. See the module level docstring for examples.
-    
+
     sv : :class:`~fse.models.sentencevectors.SentenceVectors`
         This object contains the sentence vectors inferred from the training data. There will be one such vector
         for each unique docusentence supplied during training. They may be individually accessed using the index.
-    
+
     prep : :class:`~fse.models.base_s2v.BaseSentence2VecPreparer`
         The prep object is used to transform and initialize the sv.vectors. Aditionally, it can be used
         to move the vectors to disk for training with memmap.
-    
+
     """
 
-    def __init__(self, model:BaseKeyedVectors, sv_mapfile_path:str=None, wv_mapfile_path:str=None, workers:int=1, lang_freq:str=None, **kwargs):
-        """ Average (unweighted) sentence embeddings model. Performs a simple averaging operation over all
+    def __init__(
+        self,
+        model: BaseKeyedVectors,
+        sv_mapfile_path: str = None,
+        wv_mapfile_path: str = None,
+        workers: int = 1,
+        lang_freq: str = None,
+        **kwargs
+    ):
+        """Average (unweighted) sentence embeddings model. Performs a simple averaging operation over all
         words in a sentences without further transformation.
@@ -221,15 +254,23 @@
             See https://github.com/LuminosoInsight/wordfreq
-        
+
         """
 
         super(Average, self).__init__(
-            model=model, sv_mapfile_path=sv_mapfile_path, wv_mapfile_path=wv_mapfile_path,
-            workers=workers, lang_freq=lang_freq,
-            batch_words=MAX_WORDS_IN_BATCH, batch_ngrams=MAX_NGRAMS_IN_BATCH,
-            fast_version=FAST_VERSION
-            )
+            model=model,
+            sv_mapfile_path=sv_mapfile_path,
+            wv_mapfile_path=wv_mapfile_path,
+            workers=workers,
+            lang_freq=lang_freq,
+            batch_words=MAX_WORDS_IN_BATCH,
+            batch_ngrams=MAX_NGRAMS_IN_BATCH,
+            fast_version=FAST_VERSION,
+        )
 
-    def _do_train_job(self, data_iterable:List[tuple], target:ndarray, memory:ndarray) -> [int, int]:
+    def _do_train_job(
+        self, data_iterable: List[tuple], target: ndarray, memory: ndarray
+    ) -> [int, int]:
         """ Internal routine which is called on training and performs averaging for all entries in the iterable """
-        eff_sentences, eff_words = train_average(model=self, indexed_sentences=data_iterable, target=target, memory=memory)
+        eff_sentences, eff_words = train_average(
+            model=self, indexed_sentences=data_iterable, target=target, memory=memory
+        )
         return eff_sentences, eff_words
@@ -239,3 +280,3 @@
         """ Check the sanity of all child paramters """
-        if not all(self.word_weights == 1.): 
+        if not all(self.word_weights == 1.0):
             raise ValueError("All word weights must equal one for averaging")
@@ -250,13 +291,11 @@
         pass
-    
+
     def _post_inference_calls(self, **kwargs):
-        """ Function calls to perform after training & inference
+        """Function calls to perform after training & inference
         Examples include the removal of components
         """
         pass
-    
+
     def _check_dtype_santiy(self, **kwargs):
         """ Check the dtypes of all child attributes"""
         pass
-
-    

fse/models/base_s2v.py

 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
 
-# Author: Oliver Borchers <[e-mail]>
-# Copyright (C) 2019 Oliver Borchers
+# Author: Oliver Borchers
+# Copyright (C) Oliver Borchers Oliver Borchers
 # Licensed under GNU General Public License v3.0
@@ -39,2 +39,4 @@
 
+from fse.models.utils import set_madvise_for_mmap
+
 from gensim.models.base_any2vec import BaseWordEmbeddingsModel
@@ -45,4 +47,15 @@ from gensim.models.keyedvectors import BaseKeyedVectors, FastTextKeyedVectors, _l2_norm
 
-from numpy import ndarray, memmap as np_memmap, float32 as REAL, uint32 as uINT, \
-    empty, zeros, vstack, dtype, ones, finfo, full
+from numpy import (
+    ndarray,
+    memmap as np_memmap,
+    float32 as REAL,
+    uint32 as uINT,
+    empty,
+    zeros,
+    vstack,
+    dtype,
+    ones,
+    finfo,
+    full,
+)
 
@@ -69,6 +82,17 @@ from wordfreq import available_languages, get_frequency_dict
 
+
 class BaseSentence2VecModel(SaveLoad):
-    
-    def __init__(self, model:BaseKeyedVectors, sv_mapfile_path:str=None, wv_mapfile_path:str=None, workers:int=1, lang_freq:str=None, fast_version:int=0, batch_words:int=10000, batch_ngrams:int=40, **kwargs):
-        """ Base class for all Sentence2Vec Models. Provides core functionality, such as
+    def __init__(
+        self,
+        model: BaseKeyedVectors,
+        sv_mapfile_path: str = None,
+        wv_mapfile_path: str = None,
+        workers: int = 1,
+        lang_freq: str = None,
+        fast_version: int = 0,
+        batch_words: int = 10000,
+        batch_ngrams: int = 40,
+        **kwargs,
+    ):
+        """Base class for all Sentence2Vec Models. Provides core functionality, such as
         save, load, sanity checking, frequency induction, data checking, scanning, etc.
@@ -106,10 +130,43 @@
 
+        """
+        TODO:
+
+        [ ] global:
+            [ ] windows support
+            [ ] documentation
+            [ ] more benchmarks
+            [ ] remove wv_mapfile_path?
+            [ ] modifiable sv_mapfile_path?
+
+        [ ] models:
+            [ ] check feasibility first
+            [ ] max-pooling -> easy
+            [ ] hierarchical pooling -> easy
+            [ ] discrete cosine transform -> somewhat easy, questionable
+            [ ] valve -> unclear, not cited enough
+            [ ] power-means embedding -> very large dimensionalty
+                [ ] z-score transformation is quite nice
+            
+        [ ] sentencevectors:
+            [X] similar_by_sentence model type check
+            [ ] approximate NN search for large files
+                [ ] compare ANN libraries
+                [ ] ease-of-use
+                [ ] dependencies
+                [ ] compatibility
+                [ ] memory-usage
+        """
+
+        set_madvise_for_mmap()
+
         self.workers = int(workers)
         self.batch_words = batch_words
         self.batch_ngrams = batch_ngrams
-        self.wv = None                              
-                                                    
-        self.is_ft = False                          
+        self.wv = None
 
-        self.wv_mapfile_path = Path(wv_mapfile_path) if wv_mapfile_path is not None else None
+        self.is_ft = False
+
+        self.wv_mapfile_path = (
+            Path(wv_mapfile_path) if wv_mapfile_path is not None else None
+        )
         self.wv_mapfile_shapes = {}
@@ -132,3 +189,5 @@
 
-        self.sv = SentenceVectors(vector_size=self.wv.vector_size, mapfile_path=sv_mapfile_path)
+        self.sv = SentenceVectors(
+            vector_size=self.wv.vector_size, mapfile_path=sv_mapfile_path
+        )
         self.prep = BaseSentence2VecPreparer()
@@ -139,3 +198,3 @@
     def __str__(self) -> str:
-        """ Human readable representation of the model's state.
+        """Human readable representation of the model's state.
 
@@ -150,5 +209,5 @@ Returns
 
-    def _check_and_include_model(self, model:BaseKeyedVectors):
-        """ Check if the supplied model is a compatible model. Performs all kinds of checks and small optimizations.
-        
+    def _check_and_include_model(self, model: BaseKeyedVectors):
+        """Check if the supplied model is a compatible model. Performs all kinds of checks and small optimizations.
+
         Parameters
@@ -165,7 +224,9 @@ ----------
         else:
-            raise RuntimeError(f"Model must be child of BaseWordEmbeddingsModel or BaseKeyedVectors. Received {str(model)}")
+            raise RuntimeError(
+                f"Model must be child of BaseWordEmbeddingsModel or BaseKeyedVectors. Received {str(model)}"
+            )
         self.wv.vectors_norm = None
-        
+
         if isinstance(self.wv, FastTextKeyedVectors):
-            self.wv.vectors_vocab_norm = None # Save some space
+            self.wv.vectors_vocab_norm = None  # Save some space
             self.wv.vectors_ngrams_norm = None
@@ -177,15 +238,21 @@ self.wv.vectors_vocab_norm = None
                 raise RuntimeError("FastText model requires compatible hash function")
-            if not hasattr(self.wv, 'vectors_vocab') or self.wv.vectors_vocab is None:
-                raise RuntimeError("vectors_vocab required for sentence embeddings not found.")
-            if not hasattr(self.wv, 'vectors_ngrams') or self.wv.vectors_ngrams is None:
-                raise RuntimeError("Ngram vectors required for sentence embeddings not found.")
-            
-        if not hasattr(self.wv, 'vectors') or self.wv.vectors is None:
-            raise RuntimeError("Word vectors required for sentence embeddings not found.")
-        if not hasattr(self.wv, 'vocab'):
+            if not hasattr(self.wv, "vectors_vocab") or self.wv.vectors_vocab is None:
+                raise RuntimeError(
+                    "vectors_vocab required for sentence embeddings not found."
+                )
+            if not hasattr(self.wv, "vectors_ngrams") or self.wv.vectors_ngrams is None:
+                raise RuntimeError(
+                    "Ngram vectors required for sentence embeddings not found."
+                )
+
+        if not hasattr(self.wv, "vectors") or self.wv.vectors is None:
+            raise RuntimeError(
+                "Word vectors required for sentence embeddings not found."
+            )
+        if not hasattr(self.wv, "vocab"):
             raise RuntimeError("Vocab required for sentence embeddings not found.")
-    
-    def _check_language_settings(self, lang_freq:str):
-        """ Check if the supplied language is a compatible with the wordfreq package
-        
+
+    def _check_language_settings(self, lang_freq: str):
+        """Check if the supplied language is a compatible with the wordfreq package
+
         Parameters
@@ -197,12 +264,14 @@ ----------
         """
-        if lang_freq in available_languages(wordlist='best'):
+        if lang_freq in available_languages(wordlist="best"):
             self.lang_freq = str(lang_freq)
-            logger.info("no frequency mode: using wordfreq for estimation "
-                        f"of frequency for language: {self.lang_freq}")
+            logger.info(
+                "no frequency mode: using wordfreq for estimation "
+                f"of frequency for language: {self.lang_freq}"
+            )
         else:
             raise ValueError(f"Language {lang_freq} is not available in wordfreq")
-    
-    def _induce_frequencies(self, domain:int=2**31 - 1):
-        """ Induce frequencies for a pretrained model, as not all pretrained models come with frequencies.
-        
+
+    def _induce_frequencies(self, domain: int = 2 ** 31 - 1):
+        """Induce frequencies for a pretrained model, as not all pretrained models come with frequencies.
+
         Parameters
@@ -214,3 +283,3 @@ ----------
         """
-        freq_dict = get_frequency_dict(self.lang_freq, wordlist='best')
+        freq_dict = get_frequency_dict(self.lang_freq, wordlist="best")
         for word in self.wv.index2word:
@@ -222,5 +291,5 @@ if word in freq_dict:
 
-    def _check_input_data_sanity(self, data_iterable:tuple):
-        """ Check if the input data complies with the required formats
-        
+    def _check_input_data_sanity(self, data_iterable: tuple):
+        """Check if the input data complies with the required formats
+
         Parameters
@@ -235,8 +304,10 @@ ----------
         elif isinstance(data_iterable, str):
-            raise TypeError("Passed string. Input data must be iterable list of list of tokens or tuple")
+            raise TypeError(
+                "Passed string. Input data must be iterable list of list of tokens or tuple"
+            )
         elif not hasattr(data_iterable, "__iter__"):
             raise TypeError("Iterable must provide __iter__ function")
 
-    def _log_train_end(self, eff_sentences:int, eff_words:int, overall_time:float):
-        """ Log the end of training.
+    def _log_train_end(self, eff_sentences: int, eff_words: int, overall_time: float):
+        """Log the end of training.
 
@@ -258,4 +329,6 @@ Parameters
 
-    def _check_pre_training_sanity(self, total_sentences:int, total_words:int, average_length:int, **kwargs):
-        """ Check if all available objects for training are available and compliant
+    def _check_pre_training_sanity(
+        self, total_sentences: int, total_words: int, average_length: int, **kwargs
+    ):
+        """Check if all available objects for training are available and compliant
 
@@ -272,11 +345,17 @@ Parameters
         """
-        if not hasattr(self, "wv") or self.wv is None: 
+        if not hasattr(self, "wv") or self.wv is None:
             raise RuntimeError("you must first load a valid BaseKeyedVectors object")
         if not len(self.wv.vectors):
-            raise RuntimeError("you must initialize vectors before computing sentence vectors")
+            raise RuntimeError(
+                "you must initialize vectors before computing sentence vectors"
+            )
 
         if self.is_ft and not len(self.wv.vectors_ngrams):
-            raise RuntimeError("you must initialize ngram vectors before computing sentence vectors")
+            raise RuntimeError(
+                "you must initialize ngram vectors before computing sentence vectors"
+            )
         if self.is_ft and not len(self.wv.vectors_vocab):
-            raise RuntimeError("you must initialize vectors_vocab before computing sentence vectors")
+            raise RuntimeError(
+                "you must initialize vectors_vocab before computing sentence vectors"
+            )
 
@@ -289,7 +368,7 @@ if sum([self.wv.vocab[w].count for w in self.wv.vocab]) == len(self.wv.vocab):
 
-        if not hasattr(self.sv, "vectors") or self.sv.vectors is None: 
+        if not hasattr(self.sv, "vectors") or self.sv.vectors is None:
             raise RuntimeError("initialization of Sentencevectors failed")
-        if not hasattr(self, "word_weights") or self.word_weights is None: 
+        if not hasattr(self, "word_weights") or self.word_weights is None:
             raise RuntimeError("initialization of word weights failed")
-            
+
         if not len(self.wv.vectors) == len(self.word_weights):
@@ -301,5 +380,9 @@ raise RuntimeError("Number of word vectors and weights does not match")
         if self.is_ft and self.wv.vectors_ngrams.dtype != REAL:
-            raise TypeError(f"type of wv.vectors_ngrams is wrong: {self.wv.vectors_ngrams.dtype}")
+            raise TypeError(
+                f"type of wv.vectors_ngrams is wrong: {self.wv.vectors_ngrams.dtype}"
+            )
         if self.is_ft and self.wv.vectors_vocab.dtype != REAL:
-            raise TypeError(f"type of wv.vectors_vocab is wrong: {self.wv.vectors_vocab.dtype}")
+            raise TypeError(
+                f"type of wv.vectors_vocab is wrong: {self.wv.vectors_vocab.dtype}"
+            )
         if self.sv.vectors.dtype != REAL:
@@ -310,3 +393,3 @@ raise TypeError(f"type of sv.vectors is wrong: {self.sv.vectors.dtype}")
 
-        if total_sentences is 0 or total_words is 0 or average_length is 0:
+        if total_sentences == 0 or total_words == 0 or average_length == 0:
             raise ValueError(
@@ -316,4 +399,4 @@ f"scanning the sentences returned invalid values. Check the input."
 
-    def _check_post_training_sanity(self, eff_sentences:int, eff_words:int):
-        """ Check if the training results make sense
+    def _check_post_training_sanity(self, eff_sentences: int, eff_words: int):
+        """Check if the training results make sense
 
@@ -326,12 +409,12 @@ Parameters
             Number of effective words encountered during training
-        
+
         """
-        if eff_sentences is 0 or eff_words is 0:
-            raise ValueError(
-                f"training returned invalid values. Check the input."
-            )
-    
-    def _check_indexed_sent_valid(self, iterPos:int, obj:tuple, checked:int=False) -> [int, List[str]]:
-        """ Performs a check if the passed object contains valid data
+        if eff_sentences == 0 or eff_words == 0:
+            raise ValueError(f"training returned invalid values. Check the input.")
 
+    def _check_indexed_sent_valid(
+        self, iterPos: int, obj: tuple, checked: int = False
+    ) -> [int, List[str]]:
+        """Performs a check if the passed object contains valid data
+
         Parameters
@@ -343,3 +426,3 @@ ----------
             An tuple object containing the index and sentence
-        
+
         Returns
@@ -355,3 +438,3 @@ -------
         if isinstance(obj, tuple):
-            sent = obj[0]   #Faster than obj.words
+            sent = obj[0]  # Faster than obj.words
             index = obj[1]
@@ -363,5 +446,9 @@ else:
             if not isinstance(sent, list) or not all(isinstance(w, str) for w in sent):
-                raise TypeError(f"At {iterPos}: Passed {type(sent)}: {sent}. tuple.words must contain list of str.")
+                raise TypeError(
+                    f"At {iterPos}: Passed {type(sent)}: {sent}. tuple.words must contain list of str."
+                )
             if not isinstance(index, int):
-                raise TypeError(f"At {iterPos}: Passed {type(index)}: {index}. tuple.index must contain index")
+                raise TypeError(
+                    f"At {iterPos}: Passed {type(index)}: {index}. tuple.index must contain index"
+                )
             if index < 0:
@@ -371,4 +458,4 @@ raise ValueError(f"At {iterPos}: Passed negative {index}")
 
-    def _map_all_vectors_to_disk(self, mapfile_path:Path):
-        """ Maps all vectors to disk 
+    def _map_all_vectors_to_disk(self, mapfile_path: Path):
+        """Maps all vectors to disk
 
@@ -384,11 +471,17 @@ Parameters
         self.wv_mapfile_shapes["vectors"] = self.wv.vectors.shape
-        self.wv.vectors = self._move_ndarray_to_disk(self.wv.vectors, mapfile_path=path, name="wv")
+        self.wv.vectors = self._move_ndarray_to_disk(
+            self.wv.vectors, mapfile_path=path, name="wv"
+        )
         if self.is_ft:
             self.wv_mapfile_shapes["vectors_vocab"] = self.wv.vectors_vocab.shape
             self.wv_mapfile_shapes["vectors_ngrams"] = self.wv.vectors_ngrams.shape
-            self.wv.vectors_vocab = self._move_ndarray_to_disk(self.wv.vectors_vocab, mapfile_path=self.wv_mapfile_path, name="vocab")    
-            self.wv.vectors_ngrams = self._move_ndarray_to_disk(self.wv.vectors_ngrams, mapfile_path=self.wv_mapfile_path, name="ngrams")
+            self.wv.vectors_vocab = self._move_ndarray_to_disk(
+                self.wv.vectors_vocab, mapfile_path=self.wv_mapfile_path, name="vocab"
+            )
+            self.wv.vectors_ngrams = self._move_ndarray_to_disk(
+                self.wv.vectors_ngrams, mapfile_path=self.wv_mapfile_path, name="ngrams"
+            )
 
-    def _load_all_vectors_from_disk(self, mapfile_path:Path):
-        """ Reads all vectors from disk 
+    def _load_all_vectors_from_disk(self, mapfile_path: Path):
+        """Reads all vectors from disk
 
@@ -403,12 +496,27 @@ Parameters
 
-        self.wv.vectors = np_memmap(f"{path}_wv.vectors", dtype=REAL, mode='r', shape=self.wv_mapfile_shapes["vectors"])
+        self.wv.vectors = np_memmap(
+            f"{path}_wv.vectors",
+            dtype=REAL,
+            mode="r",
+            shape=self.wv_mapfile_shapes["vectors"],
+        )
         if self.is_ft:
             self.wv.vectors_vocab = np_memmap(
-                f"{path}_vocab.vectors", dtype=REAL, mode='r', shape=self.wv_mapfile_shapes["vectors_vocab"])
+                f"{path}_vocab.vectors",
+                dtype=REAL,
+                mode="r",
+                shape=self.wv_mapfile_shapes["vectors_vocab"],
+            )
             self.wv.vectors_ngrams = np_memmap(
-                f"{path}_ngrams.vectors", dtype=REAL, mode='r', shape=self.wv_mapfile_shapes["vectors_ngrams"])
-        
-    def _move_ndarray_to_disk(self, vector:ndarray, mapfile_path:str, name:str="") -> ndarray:
-        """ Moves a numpy ndarray to disk via memmap
+                f"{path}_ngrams.vectors",
+                dtype=REAL,
+                mode="r",
+                shape=self.wv_mapfile_shapes["vectors_ngrams"],
+            )
 
+    def _move_ndarray_to_disk(
+        self, vector: ndarray, mapfile_path: str, name: str = ""
+    ) -> ndarray:
+        """Moves a numpy ndarray to disk via memmap
+
         Parameters
@@ -434,5 +542,3 @@ ----------
             logger.info(f"writing {name} to {path}")
-            memvecs = np_memmap(
-                path, dtype=REAL,
-                mode='w+', shape=shape)
+            memvecs = np_memmap(path, dtype=REAL, mode="w+", shape=shape)
             memvecs[:] = vector[:]
@@ -444,3 +550,3 @@ del memvecs, vector
 
-        readonly_memvecs = np_memmap(path, dtype=REAL, mode='r', shape=shape)
+        readonly_memvecs = np_memmap(path, dtype=REAL, mode="r", shape=shape)
         return readonly_memvecs
@@ -461,3 +567,5 @@
 
-    def _do_train_job(self, data_iterable:List[tuple], target:ndarray, memory:ndarray) -> [int, int]:
+    def _do_train_job(
+        self, data_iterable: List[tuple], target: ndarray, memory: ndarray
+    ) -> [int, int]:
         """ Function to be called on a batch of sentences. Returns eff sentences/words """
@@ -473,5 +581,5 @@ raise NotImplementedError()
         raise NotImplementedError()
-    
+
     def _post_inference_calls(self, **kwargs):
-        """ Function calls to perform after training & inference
+        """Function calls to perform after training & inference
         Examples include the removal of components
@@ -491,3 +599,3 @@ """
     def load(cls, *args, **kwargs):
-        """ Load a previously saved :class:`~fse.models.base_s2v.BaseSentence2VecModel`.
+        """Load a previously saved :class:`~fse.models.base_s2v.BaseSentence2VecModel`.
 
@@ -512,6 +620,9 @@ Parameters
         model.wv_mapfile_shapes = None
+
+        set_madvise_for_mmap()
+
         return model
 
     def save(self, *args, **kwargs):
-        """ Save the model.
+        """Save the model.
         This saved model can be loaded again using :func:`~fse.models.base_s2v.BaseSentence2VecModel.load`
@@ -533,4 +644,6 @@
 
-    def scan_sentences(self, sentences:List[tuple]=None, progress_per:int=5) -> Dict[str,int]:
-        """ Performs an initial scan of the data and reports all corresponding statistics
+    def scan_sentences(
+        self, sentences: List[tuple] = None, progress_per: int = 5
+    ) -> Dict[str, int]:
+        """Performs an initial scan of the data and reports all corresponding statistics
 
@@ -548,3 +661,3 @@ Parameters
             Dictionary containing the scan statistics
-        
+
         """
@@ -559,10 +672,16 @@ logger.info("scanning all indexed sentences and their word counts")
         max_index = 0
-        checked_sentences = 0   # We only check the first item to not constrain runtime so much
+        checked_sentences = (
+            0  # We only check the first item to not constrain runtime so much
+        )
 
         for i, obj in enumerate(sentences):
-            index, sent = self._check_indexed_sent_valid(iterPos=i, obj=obj, checked=checked_sentences)
+            index, sent = self._check_indexed_sent_valid(
+                iterPos=i, obj=obj, checked=checked_sentences
+            )
             checked_sentences += 1
             if time() - current_time > progress_per:
                 current_time = time()
-                logger.info(f"SCANNING : finished {total_sentences} sentences with {total_words} words")
+                logger.info(
+                    f"SCANNING : finished {total_sentences} sentences with {total_words} words"
+                )
 
@@ -575,3 +694,3 @@ max_index = max(max_index, index)
                 empty_sentences += 1
-        
+
         if empty_sentences:
@@ -581,3 +700,5 @@ logger.warning(f"found {empty_sentences} empty sentences")
         if max_index >= total_sentences:
-            raise RuntimeError(f"Index {max_index} is larger than number of sentences {total_sentences}")
+            raise RuntimeError(
+                f"Index {max_index} is larger than number of sentences {total_sentences}"
+            )
 
@@ -590,13 +711,15 @@ average_length = int(total_words / total_sentences)
         statistics = {
-            "total_sentences" : total_sentences,
-            "total_words" : total_words,
-            "average_length" : average_length,
-            "empty_sentences" : empty_sentences,
-            "max_index" : max_index + 1
+            "total_sentences": total_sentences,
+            "total_words": total_words,
+            "average_length": average_length,
+            "empty_sentences": empty_sentences,
+            "max_index": max_index + 1,
         }
         return statistics
-    
-    def estimate_memory(self, max_index:int, report:dict=None, **kwargs) -> Dict[str, int]:
-        """ Estimate the size of the sentence embedding
 
+    def estimate_memory(
+        self, max_index: int, report: dict = None, **kwargs
+    ) -> Dict[str, int]:
+        """Estimate the size of the sentence embedding
+
         Parameters
@@ -620,8 +743,16 @@ ----------
         report["Word Vectors"] = vocab_size * self.wv.vector_size * dtype(REAL).itemsize
-        report["Sentence Vectors"] = max_index * self.wv.vector_size * dtype(REAL).itemsize
+        report["Sentence Vectors"] = (
+            max_index * self.wv.vector_size * dtype(REAL).itemsize
+        )
         if self.is_ft:
-            report["Vocab Vectors"] = vocab_size * self.wv.vector_size * dtype(REAL).itemsize
-            report["Ngram Vectors"] = self.wv.vectors_ngrams.shape[0] * self.wv.vector_size * dtype(REAL).itemsize
+            report["Vocab Vectors"] = (
+                vocab_size * self.wv.vector_size * dtype(REAL).itemsize
+            )
+            report["Ngram Vectors"] = (
+                self.wv.vectors_ngrams.shape[0]
+                * self.wv.vector_size
+                * dtype(REAL).itemsize
+            )
         report["Total"] = sum(report.values())
-        mb_size = int(report["Total"] / 1024**2)
+        mb_size = int(report["Total"] / 1024 ** 2)
         logger.info(
@@ -633,7 +764,15 @@ f"estimated memory for {max_index} sentences with "
         if report["Total"] >= 0.95 * virtual_memory()[1]:
-            logger.warning("The embeddings will likely not fit into RAM. Consider to use mapfile_path")
+            logger.warning(
+                "The embeddings will likely not fit into RAM. Consider to use mapfile_path"
+            )
         return report
 
-    def train(self, sentences:List[tuple]=None, update:bool=False, queue_factor:int=2, report_delay:int=5) -> [int,int]:
-        """ Main routine to train an embedding. This method writes all sentences vectors into sv.vectors and is
+    def train(
+        self,
+        sentences: List[tuple] = None,
+        update: bool = False,
+        queue_factor: int = 2,
+        report_delay: int = 5,
+    ) -> [int, int]:
+        """Main routine to train an embedding. This method writes all sentences vectors into sv.vectors and is
         used for computing embeddings for large chunks of data. This method also handles post-training transformations,
@@ -665,4 +804,6 @@ such as computing the SVD of the sentence vectors.
         self.estimate_memory(**statistics)
-        self.prep.prepare_vectors(sv=self.sv, total_sentences=statistics["max_index"], update=update)
-        
+        self.prep.prepare_vectors(
+            sv=self.sv, total_sentences=statistics["max_index"], update=update
+        )
+
         # Preform post-tain calls (i.e weight computation)
@@ -676,7 +817,14 @@ self._pre_train_calls(**statistics)
 
-        _, eff_sentences, eff_words = self._train_manager(data_iterable=sentences, total_sentences=statistics["total_sentences"], queue_factor=queue_factor, report_delay=report_delay)
+        _, eff_sentences, eff_words = self._train_manager(
+            data_iterable=sentences,
+            total_sentences=statistics["total_sentences"],
+            queue_factor=queue_factor,
+            report_delay=report_delay,
+        )
 
         overall_time = time() - start_time
 
-        self._check_post_training_sanity(eff_sentences=eff_sentences, eff_words=eff_words)
+        self._check_post_training_sanity(
+            eff_sentences=eff_sentences, eff_words=eff_words
+        )
 
@@ -686,8 +834,10 @@ # Preform post-tain calls (i.e principal component removal)
 
-        self._log_train_end(eff_sentences=eff_sentences, eff_words=eff_words, overall_time=overall_time)
+        self._log_train_end(
+            eff_sentences=eff_sentences, eff_words=eff_words, overall_time=overall_time
+        )
 
-        return eff_sentences, eff_words    
+        return eff_sentences, eff_words
 
-    def infer(self, sentences:List[tuple]=None, use_norm=False) -> ndarray:
-        """ Secondary routine to train an embedding. This method is essential for small batches of sentences,
+    def infer(self, sentences: List[tuple] = None, use_norm=False) -> ndarray:
+        """Secondary routine to train an embedding. This method is essential for small batches of sentences,
         which require little computation. Note: This method does not apply post-training transformations,
@@ -715,3 +865,3 @@ only post inference calls (such as removing principal components).
         mem = self._get_thread_working_mem()
-        
+
         job_batch, batch_size = [], 0
@@ -735,5 +885,11 @@ for data_idx, data in enumerate(sentences):
 
-    def _train_manager(self, data_iterable:List[tuple], total_sentences:int=None, queue_factor:int=2, report_delay:int=5):
-        """ Manager for the multi-core implementation. Directly adapted from gensim
-        
+    def _train_manager(
+        self,
+        data_iterable: List[tuple],
+        total_sentences: int = None,
+        queue_factor: int = 2,
+        report_delay: int = 5,
+    ):
+        """Manager for the multi-core implementation. Directly adapted from gensim
+
         Parameters
@@ -756,5 +912,3 @@ ----------
         workers = [
-            threading.Thread(
-                target=self._worker_loop,
-                args=(job_queue, progress_queue))
+            threading.Thread(target=self._worker_loop, args=(job_queue, progress_queue))
             for _ in range(self.workers)
@@ -764,5 +918,3 @@ ]
         workers.append(
-            threading.Thread(
-            target=self._job_producer,
-            args=(data_iterable, job_queue))
+            threading.Thread(target=self._job_producer, args=(data_iterable, job_queue))
         )
@@ -775,4 +927,3 @@
         jobs, eff_sentences, eff_words = self._log_train_progress(
-            progress_queue, total_sentences=total_sentences,
-            report_delay=report_delay
+            progress_queue, total_sentences=total_sentences, report_delay=report_delay
         )
@@ -782,3 +933,3 @@ return jobs, eff_sentences, eff_words
     def _worker_loop(self, job_queue, progress_queue):
-        """ Train the model, lifting batches of data from the queue.
+        """Train the model, lifting batches of data from the queue.
 
@@ -807,11 +958,13 @@ This function will be called in parallel by multiple workers (threads or processes) to make
                 # no more jobs => quit this worker
-                break  
-            eff_sentences, eff_words = self._do_train_job(data_iterable=job, target=self.sv.vectors, memory=mem)
+                break
+            eff_sentences, eff_words = self._do_train_job(
+                data_iterable=job, target=self.sv.vectors, memory=mem
+            )
             progress_queue.put((len(job), eff_sentences, eff_words))
             jobs_processed += 1
         logger.debug(f"worker exiting, processed {jobs_processed} jobs")
-    
-    def _job_producer(self, data_iterable:List[tuple], job_queue:Queue):
-        """ Fill the jobs queue using the data found in the input stream.
 
+    def _job_producer(self, data_iterable: List[tuple], job_queue: Queue):
+        """Fill the jobs queue using the data found in the input stream.
+
         Each job is represented as a batch of tuple
@@ -841,3 +994,3 @@
                 job_batch, batch_size = [data], data_length
-        
+
         if job_batch:
@@ -850,6 +1003,8 @@ job_no += 1
         logger.debug(f"job loop exiting, total {job_no} jobs")
-    
-    def _log_train_progress(self, progress_queue:Queue, total_sentences:int=None, report_delay:int=5):
-        """ Log the training process after a couple of seconds.
 
+    def _log_train_progress(
+        self, progress_queue: Queue, total_sentences: int = None, report_delay: int = 5
+    ):
+        """Log the training process after a couple of seconds.
+
         Parameters
@@ -881,3 +1036,5 @@ ----------
                 unfinished_worker_count -= 1
-                logger.info(f"worker thread finished; awaiting finish of {unfinished_worker_count} more threads")
+                logger.info(
+                    f"worker thread finished; awaiting finish of {unfinished_worker_count} more threads"
+                )
                 continue
@@ -892,15 +1049,21 @@
 
-                logger.info("PROGRESS : finished {:3.2f}% with {} sentences and {} words, {} sentences/s".format(
-                    100 * (eff_sentences/total_sentences),
-                    eff_sentences, eff_words,
-                    int((eff_sentences-sentence_inc) / report_delay)
-                ))
+                logger.info(
+                    "PROGRESS : finished {:3.2f}% with {} sentences and {} words, {} sentences/s".format(
+                        100 * (eff_sentences / total_sentences),
+                        eff_sentences,
+                        eff_words,
+                        int((eff_sentences - sentence_inc) / report_delay),
+                    )
+                )
                 sentence_inc = eff_sentences
-        
+
         return jobs, eff_sentences, eff_words
 
+
 class BaseSentence2VecPreparer(SaveLoad):
     """ Contains helper functions to perpare the weights for the training of BaseSentence2VecModel """
 
-    def prepare_vectors(self, sv:SentenceVectors, total_sentences:int, update:bool=False):
+    def prepare_vectors(
+        self, sv: SentenceVectors, total_sentences: int, update: bool = False
+    ):
         """Build tables and model weights based on final vocabulary settings."""
@@ -912,3 +1075,3 @@ if not update:
 
-    def reset_vectors(self, sv:SentenceVectors, total_sentences:int):
+    def reset_vectors(self, sv: SentenceVectors, total_sentences: int):
         """Initialize all sentence vectors to zero and overwrite existing files"""
@@ -918,7 +1081,10 @@ logger.info(f"initializing sentence vectors for {total_sentences} sentences")
             sv.vectors = np_memmap(
-                str(sv.mapfile_path) + '.vectors', dtype=REAL,
-                mode='w+', shape=(total_sentences, sv.vector_size))
+                str(sv.mapfile_path) + ".vectors",
+                dtype=REAL,
+                mode="w+",
+                shape=(total_sentences, sv.vector_size),
+            )
         else:
             sv.vectors = empty((total_sentences, sv.vector_size), dtype=REAL)
-        
+
         for i in range(total_sentences):
@@ -928,3 +1094,3 @@ sv.vectors[i] = full(shape=sv.vector_size, fill_value=EPS, dtype=REAL)
 
-    def update_vectors(self, sv:SentenceVectors, total_sentences:int):
+    def update_vectors(self, sv: SentenceVectors, total_sentences: int):
         """Given existing sentence vectors, append new ones"""
@@ -937,4 +1103,7 @@ logger.info(f"appending sentence vectors for {total_sentences} sentences")
             sv.vectors = np_memmap(
-                str(sv.mapfile_path) + '.vectors', dtype=REAL,
-                mode='r+', shape=(sentences_after, sv.vector_size))
+                str(sv.mapfile_path) + ".vectors",
+                dtype=REAL,
+                mode="r+",
+                shape=(sentences_after, sv.vector_size),
+            )
             for i in range(sentences_before, sentences_after):
@@ -947,2 +1116,2 @@ sv.vectors[i] = full(shape=sv.vector_size, fill_value=EPS, dtype=REAL)
             sv.vectors = vstack([sv.vectors, newvectors])
-        sv.vectors_norm = None
+        sv.vectors_norm = None

fse/models/sentencevectors.py

 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
 
-# Author: Oliver Borchers <[e-mail]>
-# Copyright (C) 2019 Oliver Borchers
+# Author: Oliver Borchers
+# Copyright (C) Oliver Borchers Oliver Borchers
 
@@ -12,7 +12,18 @@
 
+from fse.models.utils import set_madvise_for_mmap
+
 from gensim.models.keyedvectors import BaseKeyedVectors
 
-from numpy import dot, float32 as REAL, memmap as np_memmap, \
-    double, array, zeros, vstack, sqrt, newaxis, integer, \
-    ndarray, sum as np_sum, prod, argmax
+from numpy import (
+    dot,
+    float32 as REAL,
+    memmap as np_memmap,
+    array,
+    zeros,
+    vstack,
+    sqrt,
+    newaxis,
+    integer,
+    ndarray,
+)
 
@@ -29,14 +40,17 @@ from gensim import utils, matutils
 
+
 class SentenceVectors(utils.SaveLoad):
+    def __init__(self, vector_size: int, mapfile_path: str = None):
 
-    def __init__(self, vector_size:int, mapfile_path:str=None):
-        self.vector_size = vector_size                      # Size of vectors
-        self.vectors = zeros((0, vector_size), REAL)        # Vectors for sentences
+        set_madvise_for_mmap()
+
+        self.vector_size = vector_size  # Size of vectors
+        self.vectors = zeros((0, vector_size), REAL)  # Vectors for sentences
         self.vectors_norm = None
 
         # File for numpy memmap
-        self.mapfile_path = Path(mapfile_path) if mapfile_path is not None else None            
+        self.mapfile_path = Path(mapfile_path) if mapfile_path is not None else None
         self.mapfile_shape = None
 
-    def __getitem__(self, entities:int) -> ndarray:
+    def __getitem__(self, entities: int) -> ndarray:
         """Get vector representation of `entities`.
@@ -56,3 +70,9 @@
 
-        if isinstance(entities, (int, integer,)):
+        if isinstance(
+            entities,
+            (
+                int,
+                integer,
+            ),
+        ):
             return self.get_vector(entities)
@@ -62,4 +82,10 @@
 
-    def __contains__(self, index:int) -> bool:
-        if isinstance(index, (int, integer,)):
+    def __contains__(self, index: int) -> bool:
+        if isinstance(
+            index,
+            (
+                int,
+                integer,
+            ),
+        ):
             return index < len(self)
@@ -72,6 +98,8 @@ else:
 
-    def _load_all_vectors_from_disk(self, mapfile_path:Path):
+    def _load_all_vectors_from_disk(self, mapfile_path: Path):
         """ Reads all vectors from disk """
         path = str(mapfile_path.absolute())
-        self.vectors = np_memmap(f"{path}.vectors", dtype=REAL, mode='r+', shape=self.mapfile_shape)
+        self.vectors = np_memmap(
+            f"{path}.vectors", dtype=REAL, mode="r+", shape=self.mapfile_shape
+        )
 
@@ -97,3 +125,3 @@ def save(self, *args, **kwargs):
             ignore.append("vectors")
-        kwargs['ignore'] = kwargs.get('ignore', ignore)
+        kwargs["ignore"] = kwargs.get("ignore", ignore)
         super(SentenceVectors, self).save(*args, **kwargs)
@@ -108,5 +136,6 @@
             sv._load_all_vectors_from_disk(mapfile_path=path)
+        set_madvise_for_mmap()
         return sv
 
-    def get_vector(self, index:int, use_norm:bool=False) -> ndarray:
+    def get_vector(self, index: int, use_norm: bool = False) -> ndarray:
         """Get sentence representations in vector space, as a 1D numpy array.
@@ -143,3 +172,3 @@
 
-    def init_sims(self, replace:bool=False):
+    def init_sims(self, replace: bool = False):
         """Precompute L2-normalized vectors.
@@ -152,11 +181,14 @@
         """
-        if getattr(self, 'vectors_norm', None) is None or replace:
+        if getattr(self, "vectors_norm", None) is None or replace:
             logger.info("precomputing L2-norms of sentence vectors")
             if not replace and self.mapfile_path is not None:
                 self.vectors_norm = np_memmap(
-                    self.mapfile_path + '.vectors_norm', dtype=REAL,
-                    mode='w+', shape=self.vectors.shape)
+                    self.mapfile_path + ".vectors_norm",
+                    dtype=REAL,
+                    mode="w+",
+                    shape=self.vectors.shape,
+                )
             self.vectors_norm = _l2_norm(self.vectors, replace=replace)
 
-    def similarity(self, d1:int, d2:int) -> float:
+    def similarity(self, d1: int, d2: int) -> float:
         """Compute cosine similarity between two sentences from the training set.
@@ -167,5 +199,5 @@
         d1 : int
-            index of sentence 
+            index of sentence
         d2 : int
-            index of sentence 
+            index of sentence
 
@@ -180,3 +212,3 @@ Returns
 
-    def distance(self, d1:int, d2:int) -> float:
+    def distance(self, d1: int, d2: int) -> float:
         """Compute cosine similarity between two sentences from the training set.
@@ -187,5 +219,5 @@
         d1 : int
-            index of sentence 
+            index of sentence
         d2 : int
-            index of sentence 
+            index of sentence
 
@@ -200,5 +232,10 @@ Returns
 
-    def most_similar(self, positive:[int,ndarray]=None, negative:[int,ndarray]=None, 
-                        indexable:[IndexedList,IndexedLineDocument]=None, topn:int=10, 
-                        restrict_size:[int, Tuple[int, int]]=None) -> List[Tuple[int,float]]:
+    def most_similar(
+        self,
+        positive: [int, ndarray] = None,
+        negative: [int, ndarray] = None,
+        indexable: [IndexedList, IndexedLineDocument] = None,
+        topn: int = 10,
+        restrict_size: [int, Tuple[int, int]] = None,
+    ) -> List[Tuple[int, float]]:
 
@@ -258,3 +295,3 @@ """Find the top-N most similar sentences.
         negative = [
-            (sent, -1.0) if isinstance(sent, (int, integer,  ndarray)) else sent
+            (sent, -1.0) if isinstance(sent, (int, integer, ndarray)) else sent
             for sent in negative
@@ -282,3 +319,5 @@ ]
 
-        limited = self.vectors_norm if restrict_size is None else self.vectors_norm[lo:hi]
+        limited = (
+            self.vectors_norm if restrict_size is None else self.vectors_norm[lo:hi]
+        )
         dists = dot(limited, mean)
@@ -289,11 +328,25 @@ if not topn:
         best_off = best + lo
-        
+
         if indexable is not None:
-            result = [(indexable[off_idx], off_idx, float(dists[idx])) for off_idx, idx in zip(best_off, best) if off_idx not in all_sents]
+            result = [
+                (indexable[off_idx], off_idx, float(dists[idx]))
+                for off_idx, idx in zip(best_off, best)
+                if off_idx not in all_sents
+            ]
         else:
-            result = [(off_idx, float(dists[idx])) for off_idx, idx in zip(best_off, best) if off_idx not in all_sents]
+            result = [
+                (off_idx, float(dists[idx]))
+                for off_idx, idx in zip(best_off, best)
+                if off_idx not in all_sents
+            ]
         return result[:topn]
 
-    def similar_by_word(self, word:str, wv:BaseKeyedVectors, indexable:[IndexedList,IndexedLineDocument]=None, topn:int=10, 
-                        restrict_size:[int,Tuple[int, int]]=None) -> List[Tuple[int,float]]:
+    def similar_by_word(
+        self,
+        word: str,
+        wv: BaseKeyedVectors,
+        indexable: [IndexedList, IndexedLineDocument] = None,
+        topn: int = 10,
+        restrict_size: [int, Tuple[int, int]] = None,
+    ) -> List[Tuple[int, float]]:
 
@@ -329,7 +382,18 @@ """Find the top-N most similar sentences to a given word.
         """
-        return self.most_similar(positive=wv[word], indexable=indexable, topn=topn, restrict_size=restrict_size)
+        return self.most_similar(
+            positive=wv[word],
+            indexable=indexable,
+            topn=topn,
+            restrict_size=restrict_size,
+        )
 
-    def similar_by_sentence(self, sentence:List[str], model, indexable:[IndexedList,IndexedLineDocument]=None, topn:int=10,
-                            restrict_size:[int,Tuple[int, int]]=None) -> List[Tuple[int,float]]:
-        
+    def similar_by_sentence(
+        self,
+        sentence: List[str],
+        model,
+        indexable: [IndexedList, IndexedLineDocument] = None,
+        topn: int = 10,
+        restrict_size: [int, Tuple[int, int]] = None,
+    ) -> List[Tuple[int, float]]:
+
         """Find the top-N most similar sentences to a given sentence.
@@ -364,8 +428,21 @@
         """
+        infer_op = getattr(model, "infer", None)
+        if not callable(infer_op):
+            raise RuntimeError(
+                "Model does not have infer method. Make sure to pass a BaseSentence2VecModel"
+            )
+
         vector = model.infer([(sentence, 0)])
-        return self.most_similar(positive=vector, indexable=indexable, topn=topn, restrict_size=restrict_size)
-    
-    def similar_by_vector(self, vector:ndarray, indexable:[IndexedList,IndexedLineDocument]=None, topn:int=10,
-                        restrict_size:[int,Tuple[int, int]]=None) -> List[Tuple[int,float]]:
+        return self.most_similar(
+            positive=vector, indexable=indexable, topn=topn, restrict_size=restrict_size
+        )
 
+    def similar_by_vector(
+        self,
+        vector: ndarray,
+        indexable: [IndexedList, IndexedLineDocument] = None,
+        topn: int = 10,
+        restrict_size: [int, Tuple[int, int]] = None,
+    ) -> List[Tuple[int, float]]:
+
         """Find the top-N most similar sentences to a given vector.
@@ -398,4 +475,7 @@
         """
-        return self.most_similar(positive=vector, indexable=indexable, topn=topn, restrict_size=restrict_size)
+        return self.most_similar(
+            positive=vector, indexable=indexable, topn=topn, restrict_size=restrict_size
+        )
 
+
 def _l2_norm(m, replace=False):
@@ -423,2 +503,2 @@ """Return an L2-normalized version of a matrix.
     else:
-        return (m / dist).astype(REAL)
+        return (m / dist).astype(REAL)

fse/models/sif.py

 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
 
-# Author: Oliver Borchers <[e-mail]>
-# Copyright (C) 2019 Oliver Borchers
+# Author: Oliver Borchers
+# Copyright (C) Oliver Borchers Oliver Borchers
 
@@ -12,3 +12,3 @@ from fse.models.average import Average
 
-from numpy import ndarray, float32 as REAL, ones, zeros
+from numpy import ndarray, float32 as REAL, zeros, isfinite
 
@@ -19,6 +19,16 @@ import logging
 
+
 class SIF(Average):
-
-    def __init__(self, model:BaseKeyedVectors, alpha:float=1e-3, components:int=1, sv_mapfile_path:str=None, wv_mapfile_path:str=None, workers:int=1, lang_freq:str=None):
-        """ Smooth-inverse frequency (SIF) weighted sentence embeddings model. Performs a weighted averaging operation over all
+    def __init__(
+        self,
+        model: BaseKeyedVectors,
+        alpha: float = 1e-3,
+        components: int = 1,
+        cache_size_gb: float = 1.0,
+        sv_mapfile_path: str = None,
+        wv_mapfile_path: str = None,
+        workers: int = 1,
+        lang_freq: str = None,
+    ):
+        """Smooth-inverse frequency (SIF) weighted sentence embeddings model. Performs a weighted averaging operation over all
         words in a sentences. After training, the model removes a number of singular vectors.
@@ -38,2 +48,4 @@
             Corresponds to the number of singular vectors to remove from the sentence embeddings.
+        cache_size_gb : float, optional
+            Cache size for computing the singular vectors in GB.
         sv_mapfile_path : str, optional
@@ -53,3 +65,3 @@ Optional path to store the sentence-vectors in for very large datasets. Used for memmap.
             See https://github.com/LuminosoInsight/wordfreq
-        
+
         """
@@ -59,15 +71,25 @@
         self.components = int(components)
+        self.cache_size_gb = float(cache_size_gb)
         self.svd_res = None
 
+        if lang_freq is None:
+            logger.info(
+                "make sure you are using a model with valid word-frequency information. Otherwise use lang_freq argument."
+            )
+
         super(SIF, self).__init__(
-            model=model, sv_mapfile_path=sv_mapfile_path, wv_mapfile_path=wv_mapfile_path,
-            workers=workers, lang_freq=lang_freq)
+            model=model,
+            sv_mapfile_path=sv_mapfile_path,
+            wv_mapfile_path=wv_mapfile_path,
+            workers=workers,
+            lang_freq=lang_freq,
+        )
 
     def _check_parameter_sanity(self):
         """ Check the sanity of all paramters """
-        if not all(self.word_weights <= 1.) or not all(self.word_weights >= 0.): 
+        if not all(self.word_weights <= 1.0) or not all(self.word_weights >= 0.0):
             raise ValueError("For SIF, all word weights must be 0 <= w_weight <= 1")
-        if self.alpha <= 0.:
+        if self.alpha <= 0.0:
             raise ValueError("Alpha must be greater than zero.")
-        if self.components < 0.:
+        if self.components < 0.0:
             raise ValueError("Components must be greater or equal zero")
@@ -78,8 +100,14 @@
         self._compute_sif_weights()
-    
+
     def _post_train_calls(self):
         """ Function calls to perform after training, such as computing eigenvectors """
         if self.components > 0:
-            self.svd_res = compute_principal_components(self.sv.vectors, components=self.components)
-            remove_principal_components(self.sv.vectors, svd_res=self.svd_res, inplace=True)
+            self.svd_res = compute_principal_components(
+                self.sv.vectors,
+                components=self.components,
+                cache_size_gb=self.cache_size_gb,
+            )
+            remove_principal_components(
+                self.sv.vectors, svd_res=self.svd_res, inplace=True
+            )
         else:
@@ -89,6 +117,8 @@ self.svd_res = 0
 
-    def _post_inference_calls(self, output:ndarray):
+    def _post_inference_calls(self, output: ndarray):
         """ Function calls to perform after training & inference """
         if self.svd_res is None:
-            raise RuntimeError("You must first train the model to obtain SVD components")
+            raise RuntimeError(
+                "You must first train the model to obtain SVD components"
+            )
         elif self.components > 0:
@@ -107,3 +137,5 @@ remove_principal_components(output, svd_res=self.svd_res, inplace=True)
             if self.svd_res[1].dtype != REAL:
-                raise TypeError(f"type of svd components is wrong: {self.svd_res[1].dtype}")
+                raise TypeError(
+                    f"type of svd components is wrong: {self.svd_res[1].dtype}"
+                )
 
@@ -123,2 +155,9 @@ def _compute_sif_weights(self):
 
-        self.word_weights = (self.alpha / (self.alpha + pw)).astype(REAL)
+        self.word_weights = (self.alpha / (self.alpha + pw)).astype(REAL)
+
+        if not all(isfinite(self.word_weights)) or any(self.word_weights < 0):
+            raise RuntimeError(
+                "Encountered nan values. "
+                "This likely happens because the word frequency information is wrong/missing. "
+                "Consider restarting using lang_freq argument to infer frequency. "
+            )

fse/models/usif.py

 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
 
-# Author: Oliver Borchers <[e-mail]>
-# Copyright (C) 2019 Oliver Borchers
+# Author: Oliver Borchers
+# Copyright (C) Oliver Borchers Oliver Borchers
 
@@ -12,3 +12,3 @@ from fse.models.average import Average
 
-from numpy import ndarray, float32 as REAL, zeros
+from numpy import ndarray, float32 as REAL, zeros, isfinite
 
@@ -19,6 +19,16 @@ import logging
 
+
 class uSIF(Average):
-
-    def __init__(self, model:BaseKeyedVectors, length:int=None, components:int=5, sv_mapfile_path:str=None, wv_mapfile_path:str=None, workers:int=1, lang_freq:str=None):        
-        """ Unsupervised smooth-inverse frequency (uSIF) weighted sentence embeddings model. Performs a weighted averaging operation over all
+    def __init__(
+        self,
+        model: BaseKeyedVectors,
+        length: int = None,
+        components: int = 5,
+        cache_size_gb: float = 1.0,
+        sv_mapfile_path: str = None,
+        wv_mapfile_path: str = None,
+        workers: int = 1,
+        lang_freq: str = None,
+    ):
+        """Unsupervised smooth-inverse frequency (uSIF) weighted sentence embeddings model. Performs a weighted averaging operation over all
         words in a sentences. After training, the model removes a number of weighted singular vectors.
@@ -42,2 +52,4 @@
             Is equivalent to m in the paper.
+        cache_size_gb : float, optional
+            Cache size for computing the singular vectors in GB.
         sv_mapfile_path : str, optional
@@ -57,3 +69,3 @@ Optional path to store the sentence-vectors in for very large datasets. Used for memmap.
             See https://github.com/LuminosoInsight/wordfreq
-        
+
         """
@@ -63,14 +75,24 @@
         self.components = int(components)
+        self.cache_size_gb = float(cache_size_gb)
         self.svd_res = None
         self.svd_weights = None
 
-        super(Average, self).__init__(
-            model=model, sv_mapfile_path=sv_mapfile_path, wv_mapfile_path=wv_mapfile_path,
-            workers=workers, lang_freq=lang_freq)
+        if lang_freq is None:
+            logger.info(
+                "make sure you are using a model with valid word-frequency information. Otherwise use lang_freq argument."
+            )
 
+        super(uSIF, self).__init__(
+            model=model,
+            sv_mapfile_path=sv_mapfile_path,
+            wv_mapfile_path=wv_mapfile_path,
+            workers=workers,
+            lang_freq=lang_freq,
+        )
+
     def _check_parameter_sanity(self):
         """ Check the sanity of all paramters """
-        if self.length <= 0.:
+        if self.length <= 0.0:
             raise ValueError("Length must be greater than zero.")
-        if self.components < 0.:
+        if self.components < 0.0:
             raise ValueError("Components must be greater or equal zero")
@@ -86,5 +108,16 @@
         if self.components > 0:
-            self.svd_res = compute_principal_components(self.sv.vectors, components=self.components)
-            self.svd_weights = (self.svd_res[0] ** 2) / (self.svd_res[0] ** 2).sum().astype(REAL)
-            remove_principal_components(self.sv.vectors, svd_res=self.svd_res, weights=self.svd_weights, inplace=True)
+            self.svd_res = compute_principal_components(
+                self.sv.vectors,
+                components=self.components,
+                cache_size_gb=self.cache_size_gb,
+            )
+            self.svd_weights = (self.svd_res[0] ** 2) / (
+                self.svd_res[0] ** 2
+            ).sum().astype(REAL)
+            remove_principal_components(
+                self.sv.vectors,
+                svd_res=self.svd_res,
+                weights=self.svd_weights,
+                inplace=True,
+            )
         else:
@@ -94,11 +127,15 @@ self.svd_res = 0
 
-    def _post_inference_calls(self, output:ndarray):
+    def _post_inference_calls(self, output: ndarray):
         """ Function calls to perform after training & inference """
         if self.svd_res is None:
-            raise RuntimeError("You must first train the model to obtain SVD components")
+            raise RuntimeError(
+                "You must first train the model to obtain SVD components"
+            )
         elif self.components > 0:
-            remove_principal_components(output, svd_res=self.svd_res, weights=self.svd_weights, inplace=True)
+            remove_principal_components(
+                output, svd_res=self.svd_res, weights=self.svd_weights, inplace=True
+            )
         else:
             logger.info(f"no removal of principal components")
-    
+
     def _check_dtype_santiy(self):
@@ -112,5 +149,9 @@ """ Check the dtypes of all attributes """
             if self.svd_res[1].dtype != REAL:
-                raise TypeError(f"type of svd components is wrong: {self.svd_res[1].dtype}")
+                raise TypeError(
+                    f"type of svd components is wrong: {self.svd_res[1].dtype}"
+                )
             if self.svd_weights.dtype != REAL:
-                raise TypeError(f"type of svd weights is wrong: {self.svd_weights.dtype}")
+                raise TypeError(
+                    f"type of svd weights is wrong: {self.svd_weights.dtype}"
+                )
 
@@ -130,7 +171,14 @@ def _compute_usif_weights(self):
 
-        threshold = 1 - (1-(1/v)) ** self.length
+        threshold = 1 - (1 - (1 / v)) ** self.length
         alpha = sum(pw > threshold) / v
-        z = v/2
-        a = (1 - alpha)/(alpha * z)
+        z = v / 2
+        a = (1 - alpha) / (alpha * z)
 
-        self.word_weights = (a / ((a/2) + pw)).astype(REAL)
+        self.word_weights = (a / ((a / 2) + pw)).astype(REAL)
+
+        if not all(isfinite(self.word_weights)):
+            raise RuntimeError(
+                "Encountered nan values. "
+                "This likely happens because the word frequency information is wrong/missing. "
+                "Consider restarting using lang_freq argument to infer frequency. "
+            )

fse/models/utils.py

 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
 
-# Author: Oliver Borchers <[e-mail]>
-# Copyright (C) 2019 Oliver Borchers
+# Author: Oliver Borchers
+# Copyright (C) Oliver Borchers Oliver Borchers
 
 from sklearn.decomposition import TruncatedSVD
 
-from numpy import ndarray, float32 as REAL, ones
+from numpy import ndarray, float32 as REAL, ones, vstack, dtype
+from numpy.random import choice
 
@@ -15,9 +16,46 @@ from time import time
 
+from sys import platform
+
+import ctypes
+
 logger = logging.getLogger(__name__)
 
-def compute_principal_components(vectors:ndarray, components:int=1) -> ndarray:
-    """ Method used to compute the first singular vectors of a given matrix
 
+def set_madvise_for_mmap(return_madvise: bool = False) -> object:
+    """Method used to set madvise parameters.
+    This problem adresses the memmap issue raised in https://github.com/numpy/numpy/issues/13172
+    The issue is not applicable for windows
+
     Parameters
     ----------
+    return_madvise : bool
+        Returns the madvise object for unittests, se test_utils.py
+
+    Returns
+    -------
+    object
+        madvise object
+
+    """
+
+    if platform in ["linux", "linux2", "darwin", "aix"]:
+        if platform == "darwin":
+            # Path different for Macos
+            madvise = ctypes.CDLL("libc.dylib").madvise
+        if platform in ["linux", "linux2", "aix"]:
+            madvise = ctypes.CDLL("libc.so.6").madvise
+        madvise.argtypes = [ctypes.c_void_p, ctypes.c_size_t, ctypes.c_int]
+        madvise.restype = ctypes.c_int
+
+        if return_madvise:
+            return madvise
+
+
+def compute_principal_components(
+    vectors: ndarray, components: int = 1, cache_size_gb: float = 1.0
+) -> [ndarray, ndarray]:
+    """Method used to compute the first singular vectors of a given (sub)matrix
+
+    Parameters
+    ----------
     vectors : ndarray
@@ -27,2 +65,4 @@ (Sentence) vectors to compute the truncated SVD on
         Number of singular values/vectors to compute
+    cache_size_gb : float, optional
+            Cache size for computing the principal components in GB
 
@@ -35,11 +75,33 @@ Returns
     start = time()
-    svd = TruncatedSVD(n_components=components, n_iter=7, random_state=42, algorithm="randomized")
-    svd.fit(vectors)
+    num_vectors = vectors.shape[0]
+    svd = TruncatedSVD(
+        n_components=components, n_iter=7, random_state=42, algorithm="randomized"
+    )
+
+    sample_size = int(
+        1024 ** 3 * cache_size_gb / (vectors.shape[1] * dtype(REAL).itemsize)
+    )
+
+    if sample_size > num_vectors:
+        svd.fit(vectors)
+    else:
+        logger.info(f"sampling {sample_size} vectors to compute principal components")
+        sample_indices = choice(range(num_vectors), replace=False, size=int(1e6))
+        svd.fit(vectors[sample_indices, :])
+
     elapsed = time()
-    logger.info(f"computing {components} principal components took {int(elapsed-start)}s")
+    logger.info(
+        f"computing {components} principal components took {int(elapsed-start)}s"
+    )
     return svd.singular_values_.astype(REAL), svd.components_.astype(REAL)
 
-def remove_principal_components(vectors:ndarray, svd_res:[ndarray, ndarray], weights:ndarray=None, inplace:bool=True) -> ndarray:
-    """ Method used to remove the first singular vectors of a given matrix
 
+def remove_principal_components(
+    vectors: ndarray,
+    svd_res: [ndarray, ndarray],
+    weights: ndarray = None,
+    inplace: bool = True,
+) -> ndarray:
+    """Method used to remove the first singular vectors of a given matrix
+
     Parameters
@@ -54,3 +116,3 @@ ----------
     inplace : bool, optional
-        If true, removes the componentens from the vectors inplace (memory efficient)
+        If true, removes the components from the vectors inplace (memory efficient)
 
@@ -62,3 +124,2 @@ Returns
     """
-    singular_values = svd_res[0].astype(REAL)
     components = svd_res[1].astype(REAL)
@@ -73,5 +134,5 @@
     output = None
-    if len(components)==1:
+    if len(components) == 1:
         if not inplace:
-            output = vectors.dot(w_comp.transpose()) * w_comp
+            output = vectors - vectors.dot(w_comp.transpose()) * w_comp
         else:
@@ -81,8 +142,11 @@ vectors -= vectors.dot(w_comp.transpose()) * w_comp
         if not inplace:
-            output = vectors.dot(w_comp.transpose()).dot(w_comp)
+            output = vectors - vectors.dot(w_comp.transpose()).dot(w_comp)
         else:
             vectors -= vectors.dot(w_comp.transpose()).dot(w_comp)
     elapsed = time()
-    logger.info(f"removing {len(components)} principal components took {int(elapsed-start)}s")
+
+    logger.info(
+        f"removing {len(components)} principal components took {int(elapsed-start)}s"
+    )
     if not inplace:
-        return output
+        return output

fse/test/test_average.py

 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
 
-# Author: Oliver Borchers <[e-mail]>
-# Copyright (C) 2019 Oliver Borchers
+# Author: Oliver Borchers
+# Copyright (C) Oliver Borchers Oliver Borchers
 
@@ -30,16 +30,31 @@ """
 W2V.build_vocab(SENTENCES)
-W2V.wv.vectors[:,] = np.arange(len(W2V.wv.vectors), dtype=np.float32)[:, None]
+W2V.wv.vectors[:,] = np.arange(
+    len(W2V.wv.vectors), dtype=np.float32
+)[:, None]
 
+
 class TestAverageFunctions(unittest.TestCase):
     def setUp(self):
-        self.sentences = [["They", "admit"], ["So", "Apple", "bought", "buds"], ["go", "12345"], ["pull", "12345678910111213"]]
-        self.sentences = [(s, i) for i,s in enumerate(self.sentences)]
+        self.sentences = [
+            ["They", "admit"],
+            ["So", "Apple", "bought", "buds"],
+            ["go", "12345"],
+            ["pull", "12345678910111213"],
+        ]
+        self.sentences = [(s, i) for i, s in enumerate(self.sentences)]
         self.model = Average(W2V)
-        self.model.prep.prepare_vectors(sv=self.model.sv, total_sentences=len(self.sentences), update=False)
+        self.model.prep.prepare_vectors(
+            sv=self.model.sv, total_sentences=len(self.sentences), update=False
+        )
         self.model._pre_train_calls()
 
     def test_cython(self):
-        from fse.models.average_inner import FAST_VERSION, MAX_WORDS_IN_BATCH, MAX_NGRAMS_IN_BATCH
+        from fse.models.average_inner import (
+            FAST_VERSION,
+            MAX_WORDS_IN_BATCH,
+            MAX_NGRAMS_IN_BATCH,
+        )
+
         self.assertTrue(FAST_VERSION)
-        self.assertEqual(10000,MAX_WORDS_IN_BATCH)
+        self.assertEqual(10000, MAX_WORDS_IN_BATCH)
         self.assertEqual(40, MAX_NGRAMS_IN_BATCH)
@@ -50,3 +65,5 @@
         mem = self.model._get_thread_working_mem()
-        output = train_average_np(self.model, self.sentences, self.model.sv.vectors, mem)
+        output = train_average_np(
+            self.model, self.sentences, self.model.sv.vectors, mem
+        )
         self.assertEqual((4, 7), output)
@@ -56,3 +73,3 @@ self.assertTrue((183 == self.model.sv[0]).all())
         self.assertTrue((self.model.wv.vocab["go"].index == self.model.sv[2]).all())
-    
+
     def test_average_train_cy_w2v(self):
@@ -63,3 +80,6 @@ self.model.sv.vectors = np.zeros_like(self.model.sv.vectors, dtype=np.float32)
         from fse.models.average_inner import train_average_cy
-        output = train_average_cy(self.model, self.sentences, self.model.sv.vectors, mem)
+
+        output = train_average_cy(
+            self.model, self.sentences, self.model.sv.vectors, mem
+        )
         self.assertEqual((4, 7), output)
@@ -74,3 +94,5 @@ self.assertTrue((183 == self.model.sv[0]).all())
         m = Average(ft)
-        m.prep.prepare_vectors(sv=m.sv, total_sentences=len(self.sentences), update=False)
+        m.prep.prepare_vectors(
+            sv=m.sv, total_sentences=len(self.sentences), update=False
+        )
         m._pre_train_calls()
@@ -82,3 +104,3 @@ m.wv.vectors = m.wv.vectors_vocab = np.ones_like(m.wv.vectors, dtype=np.float32)
         self.assertEqual((4, 10), output)
-        self.assertTrue((1. == m.sv[0]).all())
+        self.assertTrue((1.0 == m.sv[0]).all())
         self.assertTrue((1.5 == m.sv[2]).all())
@@ -94,3 +116,5 @@ self.assertTrue((2 == m.sv[3]).all())
         m = Average(ft)
-        m.prep.prepare_vectors(sv=m.sv, total_sentences=len(self.sentences), update=False)
+        m.prep.prepare_vectors(
+            sv=m.sv, total_sentences=len(self.sentences), update=False
+        )
         m._pre_train_calls()
@@ -102,5 +126,6 @@ m.wv.vectors = m.wv.vectors_vocab = np.ones_like(m.wv.vectors, dtype=np.float32)
         from fse.models.average_inner import train_average_cy
+
         output = train_average_cy(m, self.sentences, m.sv.vectors, mem)
         self.assertEqual((4, 10), output)
-        self.assertTrue((1.+EPS == m.sv[0]).all())
+        self.assertTrue((1.0 + EPS == m.sv[0]).all())
         self.assertTrue(np.allclose(1.5, m.sv[2]))
@@ -111,3 +136,5 @@ self.assertTrue(np.allclose(2, m.sv[3]))
         m1 = Average(W2V)
-        m1.prep.prepare_vectors(sv=m1.sv, total_sentences=len(self.sentences), update=False)
+        m1.prep.prepare_vectors(
+            sv=m1.sv, total_sentences=len(self.sentences), update=False
+        )
         m1._pre_train_calls()
@@ -118,3 +145,5 @@ mem1 = m1._get_thread_working_mem()
         m2 = Average(W2V)
-        m2.prep.prepare_vectors(sv=m2.sv, total_sentences=len(self.sentences), update=False)
+        m2.prep.prepare_vectors(
+            sv=m2.sv, total_sentences=len(self.sentences), update=False
+        )
         m2._pre_train_calls()
@@ -124,2 +153,3 @@ mem2 = m2._get_thread_working_mem()
         from fse.models.average_inner import train_average_cy
+
         o2 = train_average_cy(m2, self.sentences, m2.sv.vectors, mem2)
@@ -136,3 +166,5 @@
         m1 = Average(w2v)
-        m1.prep.prepare_vectors(sv=m1.sv, total_sentences=len(self.sentences), update=False)
+        m1.prep.prepare_vectors(
+            sv=m1.sv, total_sentences=len(self.sentences), update=False
+        )
         m1._pre_train_calls()
@@ -143,3 +175,5 @@ mem1 = m1._get_thread_working_mem()
         m2 = Average(w2v)
-        m2.prep.prepare_vectors(sv=m2.sv, total_sentences=len(self.sentences), update=False)
+        m2.prep.prepare_vectors(
+            sv=m2.sv, total_sentences=len(self.sentences), update=False
+        )
         m2._pre_train_calls()
@@ -149,2 +183,3 @@ mem2 = m2._get_thread_working_mem()
         from fse.models.average_inner import train_average_cy
+
         o2 = train_average_cy(m2, self.sentences, m2.sv.vectors, mem2)
@@ -159,6 +194,9 @@
         m1 = Average(ft)
-        m1.prep.prepare_vectors(sv=m1.sv, total_sentences=len(self.sentences), update=False)
+        m1.prep.prepare_vectors(
+            sv=m1.sv, total_sentences=len(self.sentences), update=False
+        )
         m1._pre_train_calls()
 
         from fse.models.average_inner import MAX_NGRAMS_IN_BATCH
+
         m1.batch_ngrams = MAX_NGRAMS_IN_BATCH
@@ -169,3 +207,5 @@ mem1 = m1._get_thread_working_mem()
         m2 = Average(ft)
-        m2.prep.prepare_vectors(sv=m2.sv, total_sentences=len(self.sentences), update=False)
+        m2.prep.prepare_vectors(
+            sv=m2.sv, total_sentences=len(self.sentences), update=False
+        )
         m2._pre_train_calls()
@@ -175,2 +215,3 @@ mem2 = m2._get_thread_working_mem()
         from fse.models.average_inner import train_average_cy
+
         o2 = train_average_cy(m2, self.sentences[:2], m2.sv.vectors, mem2)
@@ -182,13 +223,21 @@
     def test_do_train_job(self):
-        self.model.prep.prepare_vectors(sv=self.model.sv, total_sentences=len(SENTENCES), update=True)
+        self.model.prep.prepare_vectors(
+            sv=self.model.sv, total_sentences=len(SENTENCES), update=True
+        )
         mem = self.model._get_thread_working_mem()
-        self.assertEqual((100,1450), self.model._do_train_job(
-            [(s, i) for i,s in enumerate(SENTENCES)],
-            target=self.model.sv.vectors, memory=mem)
+        self.assertEqual(
+            (100, 1450),
+            self.model._do_train_job(
+                [(s, i) for i, s in enumerate(SENTENCES)],
+                target=self.model.sv.vectors,
+                memory=mem,
+            ),
         )
-        self.assertEqual((104,DIM), self.model.sv.vectors.shape)
+        self.assertEqual((104, DIM), self.model.sv.vectors.shape)
 
     def test_train(self):
-        self.assertEqual((100,1450), self.model.train([(s, i) for i,s in enumerate(SENTENCES)]))
-    
+        self.assertEqual(
+            (100, 1450), self.model.train([(s, i) for i, s in enumerate(SENTENCES)])
+        )
+
     def test_train_single_from_disk(self):
@@ -200,5 +249,7 @@ p = Path("fse/test/test_data/test_vecs")
         se1 = Average(W2V)
-        se2 = Average(W2V, sv_mapfile_path=str(p.absolute()) ,wv_mapfile_path=str(p.absolute()))
-        se1.train([(s, i) for i,s in enumerate(SENTENCES)])
-        se2.train([(s, i) for i,s in enumerate(SENTENCES)])
+        se2 = Average(
+            W2V, sv_mapfile_path=str(p.absolute()), wv_mapfile_path=str(p.absolute())
+        )
+        se1.train([(s, i) for i, s in enumerate(SENTENCES)])
+        se2.train([(s, i) for i, s in enumerate(SENTENCES)])
 
@@ -219,5 +270,10 @@ self.assertTrue(p_target.exists())
         se1 = Average(W2V, workers=2)
-        se2 = Average(W2V, workers=2, sv_mapfile_path=str(p.absolute()) ,wv_mapfile_path=str(p.absolute()))
-        se1.train([(s, i) for i,s in enumerate(SENTENCES)])
-        se2.train([(s, i) for i,s in enumerate(SENTENCES)])
+        se2 = Average(
+            W2V,
+            workers=2,
+            sv_mapfile_path=str(p.absolute()),
+            wv_mapfile_path=str(p.absolute()),
+        )
+        se1.train([(s, i) for i, s in enumerate(SENTENCES)])
+        se2.train([(s, i) for i, s in enumerate(SENTENCES)])
 
@@ -234,8 +290,11 @@ self.assertTrue(p_target.exists())
         se = Average(W2V)
-        se.word_weights = np.full(20, 2., dtype=np.float32)
+        se.word_weights = np.full(20, 2.0, dtype=np.float32)
         with self.assertRaises(ValueError):
             se._check_parameter_sanity()
 
-if __name__ == '__main__':
-    logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s', level=logging.DEBUG)
-    unittest.main()
+
+if __name__ == "__main__":
+    logging.basicConfig(
+        format="%(asctime)s : %(levelname)s : %(message)s", level=logging.DEBUG
+    )
+    unittest.main()

fse/test/test_base_s2v.py

 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
 
-# Author: Oliver Borchers <[e-mail]>
-# Copyright (C) 2019 Oliver Borchers
+# Author: Oliver Borchers
+# Copyright (C) Oliver Borchers Oliver Borchers
 
@@ -79,6 +79,6 @@ """
         self.assertTrue(isinstance(se.wv, BaseKeyedVectors))
-    
+
     def test_model_w_language(self):
         se = BaseSentence2VecModel(W2V, lang_freq="en")
-        freq = int((2**31 - 1) * get_frequency_dict("en", wordlist="best")["help"])
+        freq = int((2 ** 31 - 1) * get_frequency_dict("en", wordlist="best")["help"])
         self.assertEqual(freq, se.wv.vocab["help"].count)
@@ -153,3 +153,3 @@ self.assertEqual(21, se.wv.vocab["79"].count)
 
-        class BadIterator():
+        class BadIterator:
             def __init__(self):
@@ -161,7 +161,7 @@ pass
         with self.assertRaises(TypeError):
-            se._check_input_data_sanity(data_iterable = None)
+            se._check_input_data_sanity(data_iterable=None)
         with self.assertRaises(TypeError):
-            se._check_input_data_sanity(data_iterable = "Hello there!")
+            se._check_input_data_sanity(data_iterable="Hello there!")
         with self.assertRaises(TypeError):
-            se._check_input_data_sanity(data_iterable = BadIterator())
+            se._check_input_data_sanity(data_iterable=BadIterator())
 
@@ -175,7 +175,9 @@ def test_scan_w_list(self):
         output = str(BaseSentence2VecModel(W2V))
-        self.assertEqual("BaseSentence2VecModel based on Word2VecKeyedVectors, size=0", output)
+        self.assertEqual(
+            "BaseSentence2VecModel based on Word2VecKeyedVectors, size=0", output
+        )
 
     def test_scan_w_ituple(self):
         se = BaseSentence2VecModel(W2V)
-        id_sent = [(s, i) for i,s in enumerate(SENTENCES)]
+        id_sent = [(s, i) for i, s in enumerate(SENTENCES)]
         stats = se.scan_sentences(id_sent, progress_per=0)
@@ -191,3 +193,3 @@
         se = BaseSentence2VecModel(W2V)
-        id_sent = [(s, str(i)) for i,s in enumerate(SENTENCES)]
+        id_sent = [(s, str(i)) for i, s in enumerate(SENTENCES)]
         with self.assertRaises(TypeError):
@@ -200,3 +202,8 @@ se.scan_sentences(id_sent)
             SENTENCES[i] = []
-        self.assertEqual(3, se.scan_sentences([(s, i) for i,s in enumerate(SENTENCES)])["empty_sentences"])
+        self.assertEqual(
+            3,
+            se.scan_sentences([(s, i) for i, s in enumerate(SENTENCES)])[
+                "empty_sentences"
+            ],
+        )
 
@@ -206,7 +213,7 @@ def test_scan_w_wrong_input(self):
         sentences = ["the dog hit the car", "he was very fast"]
-        
+
         with self.assertRaises(TypeError):
             se.scan_sentences(sentences)
         with self.assertRaises(TypeError):
-            se.scan_sentences([(s, i) for i,s in enumerate(sentences)])
+            se.scan_sentences([(s, i) for i, s in enumerate(sentences)])
         with self.assertRaises(TypeError):
@@ -216,9 +223,11 @@ se.scan_sentences([list(range(10) for _ in range(2))])
         with self.assertRaises(RuntimeError):
-            se.scan_sentences([(s, i+1) for i,s in enumerate(SENTENCES)])
+            se.scan_sentences([(s, i + 1) for i, s in enumerate(SENTENCES)])
         with self.assertRaises(ValueError):
-            se.scan_sentences([(s, i-1) for i,s in enumerate(SENTENCES)])
-        
+            se.scan_sentences([(s, i - 1) for i, s in enumerate(SENTENCES)])
+
     def test_scan_w_many_to_one_input(self):
         se = BaseSentence2VecModel(W2V)
-        output = se.scan_sentences([(s, 0) for i,s in enumerate(SENTENCES)])["max_index"]
+        output = se.scan_sentences([(s, 0) for i, s in enumerate(SENTENCES)])[
+            "max_index"
+        ]
         self.assertEqual(1, output)
@@ -235,3 +244,3 @@
         with self.assertRaises(NotImplementedError):
-            se.train([(s, i) for i,s in enumerate(SENTENCES)])
+            se.train([(s, i) for i, s in enumerate(SENTENCES)])
 
@@ -254,5 +263,5 @@ def test_log_end(self):
         with self.assertRaises(NotImplementedError):
-            se._check_dtype_santiy()  
+            se._check_dtype_santiy()
         with self.assertRaises(NotImplementedError):
-            se._post_inference_calls()  
+            se._post_inference_calls()
 
@@ -265,3 +274,3 @@ def test_check_pre_train_san_no_wv(self):
         with self.assertRaises(RuntimeError):
-            se._check_pre_training_sanity(1,1,1)
+            se._check_pre_training_sanity(1, 1, 1)
 
@@ -274,3 +283,3 @@ def test_check_pre_train_san_no_wv_len(self):
         with self.assertRaises(RuntimeError):
-            se._check_pre_training_sanity(1,1,1)
+            se._check_pre_training_sanity(1, 1, 1)
 
@@ -283,7 +292,7 @@ def test_check_pre_train_san_no_ngrams_vectors(self):
         with self.assertRaises(RuntimeError):
-            se._check_pre_training_sanity(1,1,1)
+            se._check_pre_training_sanity(1, 1, 1)
         se.wv.vectors_ngrams = [1]
         se.wv.vectors_vocab = []
         with self.assertRaises(RuntimeError):
-            se._check_pre_training_sanity(1,1,1)
+            se._check_pre_training_sanity(1, 1, 1)
 
@@ -296,3 +305,3 @@ def test_check_pre_train_san_no_sv_vecs(self):
         with self.assertRaises(RuntimeError):
-            se._check_pre_training_sanity(1,1,1)
+            se._check_pre_training_sanity(1, 1, 1)
 
@@ -305,3 +314,3 @@ def test_check_pre_train_san_no_word_weights(self):
         with self.assertRaises(RuntimeError):
-            se._check_pre_training_sanity(1,1,1)
+            se._check_pre_training_sanity(1, 1, 1)
 
@@ -314,3 +323,3 @@ def test_check_pre_train_san_incos_len(self):
         with self.assertRaises(RuntimeError):
-            se._check_pre_training_sanity(1,1,1)
+            se._check_pre_training_sanity(1, 1, 1)
 
@@ -322,10 +331,10 @@ def test_check_pre_train_dtypes(self):
 
-        se.wv.vectors = np.zeros((len(se.wv.vocab),20), dtype=np.float64)
+        se.wv.vectors = np.zeros((len(se.wv.vocab), 20), dtype=np.float64)
         with self.assertRaises(TypeError):
-            se._check_pre_training_sanity(1,1,1)
-        se.wv.vectors = np.zeros((len(se.wv.vocab),20), dtype=np.float32)
+            se._check_pre_training_sanity(1, 1, 1)
+        se.wv.vectors = np.zeros((len(se.wv.vocab), 20), dtype=np.float32)
 
         se.wv.vectors_ngrams = np.ones(len(se.wv.vocab), dtype=np.float16)
         with self.assertRaises(TypeError):
-            se._check_pre_training_sanity(1,1,1)
+            se._check_pre_training_sanity(1, 1, 1)
         se.wv.vectors_ngrams = np.ones(len(se.wv.vocab), dtype=np.float32)
@@ -335,13 +344,13 @@
         with self.assertRaises(TypeError):
-            se._check_pre_training_sanity(1,1,1)
+            se._check_pre_training_sanity(1, 1, 1)
         se.wv.vectors_vocab = np.ones(len(se.wv.vocab), dtype=np.float32)
 
-        se.sv.vectors = np.zeros((len(se.wv.vocab),20), dtype=int)
+        se.sv.vectors = np.zeros((len(se.wv.vocab), 20), dtype=int)
         with self.assertRaises(TypeError):
-            se._check_pre_training_sanity(1,1,1)
-        se.sv.vectors = np.zeros((len(se.wv.vocab),20), dtype=np.float32)
+            se._check_pre_training_sanity(1, 1, 1)
+        se.sv.vectors = np.zeros((len(se.wv.vocab), 20), dtype=np.float32)
 
         se.word_weights = np.ones(len(se.wv.vocab), dtype=bool)
         with self.assertRaises(TypeError):
-            se._check_pre_training_sanity(1,1,1)
+            se._check_pre_training_sanity(1, 1, 1)
         se.word_weights = np.ones(len(se.wv.vocab), dtype=np.float32)
@@ -358,12 +367,11 @@
         # Just throws multiple warnings warning
-        se._check_pre_training_sanity(1,1,1)
-        
+        se._check_pre_training_sanity(1, 1, 1)
+
         with self.assertRaises(ValueError):
-            se._check_pre_training_sanity(0,1,1)
+            se._check_pre_training_sanity(0, 1, 1)
         with self.assertRaises(ValueError):
-            se._check_pre_training_sanity(1,0,1)
+            se._check_pre_training_sanity(1, 0, 1)
         with self.assertRaises(ValueError):
-            se._check_pre_training_sanity(1,1,0)
+            se._check_pre_training_sanity(1, 1, 0)
 
-
     def test_post_training_sanity(self):
@@ -375,6 +383,6 @@ w2v = Word2Vec()
         with self.assertRaises(ValueError):
-            se._check_post_training_sanity(0,1)
+            se._check_post_training_sanity(0, 1)
         with self.assertRaises(ValueError):
-            se._check_post_training_sanity(1,0)
-            
+            se._check_post_training_sanity(1, 0)
+
     def test_move_ndarray_to_disk_w2v(self):
@@ -384,5 +392,7 @@ se = BaseSentence2VecModel(W2V)
         p_target = Path("fse/test/test_data/test_vecs_wv.vectors")
-        se.wv.vectors[0,1] = 10
+        se.wv.vectors[0, 1] = 10
         vecs = se.wv.vectors.copy()
-        output = se._move_ndarray_to_disk(se.wv.vectors, name="wv", mapfile_path=str(p.absolute()))
+        output = se._move_ndarray_to_disk(
+            se.wv.vectors, name="wv", mapfile_path=str(p.absolute())
+        )
         self.assertTrue(p_target.exists())
@@ -429,11 +439,18 @@ self.assertFalse(output.flags.writeable)
         se = BaseSentence2VecModel(W2V, workers=2)
+
         def temp_train_job(data_iterable, target, memory):
             v1 = v2 = sum(1 for _ in data_iterable)
-            return v1*2, v2*3
+            return v1 * 2, v2 * 3
+
         se._do_train_job = temp_train_job
-        job_output = se._train_manager(data_iterable=[(s, i) for i,s in enumerate(SENTENCES)], total_sentences=len(SENTENCES),report_delay=0.01)
-        self.assertEqual((100,200,300), job_output)
+        job_output = se._train_manager(
+            data_iterable=[(s, i) for i, s in enumerate(SENTENCES)],
+            total_sentences=len(SENTENCES),
+            report_delay=0.01,
+        )
+        self.assertEqual((100, 200, 300), job_output)
 
     def test_infer_method(self):
         se = BaseSentence2VecModel(W2V)
+
         def temp_train_job(data_iterable, target, memory):
@@ -444,6 +461,8 @@ for i in data_iterable:
 
-        def pass_method(**kwargs): pass
+        def pass_method(**kwargs):
+            pass
+
         se._post_inference_calls = pass_method
         se._do_train_job = temp_train_job
-        output = se.infer([(s, i) for i,s in enumerate(SENTENCES)])
+        output = se.infer([(s, i) for i, s in enumerate(SENTENCES)])
         self.assertTrue((100 == output).all())
@@ -453,10 +472,15 @@
         se = BaseSentence2VecModel(W2V)
-        
+
         from fse.models.average_inner import MAX_WORDS_IN_BATCH
         from fse.models.average_inner import train_average_cy
+
         def _do_train_job(data_iterable, target, memory):
-            eff_sentences, eff_words = train_average_cy(model=se, indexed_sentences=data_iterable, target=target, memory=memory)
+            eff_sentences, eff_words = train_average_cy(
+                model=se, indexed_sentences=data_iterable, target=target, memory=memory
+            )
             return eff_sentences, eff_words
 
-        def pass_method(**kwargs): pass
+        def pass_method(**kwargs):
+            pass
+
         se._post_inference_calls = pass_method
@@ -473,3 +497,3 @@ se._do_train_job = _do_train_job
             bs += len(s)
-        sents = [(s, i) for i,s in enumerate(tmp)]
+        sents = [(s, i) for i, s in enumerate(tmp)]
         output = se.infer(sents)
@@ -481,2 +505,3 @@ output = output[i:]
         se = BaseSentence2VecModel(W2V)
+
         def temp_train_job(data_iterable, target, memory):
@@ -486,6 +511,9 @@ for i in data_iterable:
             return target
-        def pass_method(**kwargs): pass
+
+        def pass_method(**kwargs):
+            pass
+
         se._post_inference_calls = pass_method
         se._do_train_job = temp_train_job
-        output = se.infer([(s, 0) for i,s in enumerate(SENTENCES)])
+        output = se.infer([(s, 0) for i, s in enumerate(SENTENCES)])
         self.assertTrue((100 == output).all())
@@ -496,2 +524,3 @@ self.assertEqual((1, 5), output.shape)
         se = BaseSentence2VecModel(W2V)
+
         def temp_train_job(data_iterable, target, memory):
@@ -501,11 +530,14 @@ for i in data_iterable:
             return target
-        def pass_method(**kwargs): pass
+
+        def pass_method(**kwargs):
+            pass
+
         se._post_inference_calls = pass_method
         se._do_train_job = temp_train_job
-        output = se.infer([(s, i) for i,s in enumerate(SENTENCES)], use_norm=True)
+        output = se.infer([(s, i) for i, s in enumerate(SENTENCES)], use_norm=True)
 
-        self.assertTrue(np.allclose(1., np.sqrt(np.sum(output[0]**2))))
+        self.assertTrue(np.allclose(1.0, np.sqrt(np.sum(output[0] ** 2))))
 
+
 class TestBaseSentence2VecPreparerFunctions(unittest.TestCase):
-
     def test_reset_vectors(self):
@@ -515,3 +547,3 @@ se = BaseSentence2VecModel(W2V)
         trainables.reset_vectors(se.sv, 20)
-        self.assertEqual((20,DIM), se.sv.vectors.shape)
+        self.assertEqual((20, DIM), se.sv.vectors.shape)
         self.assertEqual(np.float32, se.sv.vectors.dtype)
@@ -528,3 +560,3 @@ self.assertTrue((EPS == se.sv.vectors).all())
         self.assertTrue(p_target.exists())
-        self.assertEqual((20,DIM), se.sv.vectors.shape)
+        self.assertEqual((20, DIM), se.sv.vectors.shape)
         self.assertEqual(np.float32, se.sv.vectors.dtype)
@@ -539,5 +571,5 @@ self.assertTrue((EPS == se.sv.vectors).all())
         trainables.reset_vectors(se.sv, 20)
-        se.sv.vectors[:] = 1.
+        se.sv.vectors[:] = 1.0
         trainables.update_vectors(se.sv, 10)
-        self.assertEqual((30,DIM), se.sv.vectors.shape)
+        self.assertEqual((30, DIM), se.sv.vectors.shape)
         self.assertEqual(np.float32, se.sv.vectors.dtype)
@@ -554,6 +586,6 @@ self.assertTrue((np.ones((20, DIM)) == se.sv.vectors[:20]).all())
         trainables.reset_vectors(se.sv, 20)
-        se.sv.vectors[:] = 1.
+        se.sv.vectors[:] = 1.0
         trainables.update_vectors(se.sv, 10)
         self.assertTrue(p_target.exists())
-        self.assertEqual((30,DIM), se.sv.vectors.shape)
+        self.assertEqual((30, DIM), se.sv.vectors.shape)
         self.assertEqual(np.float32, se.sv.vectors.dtype)
@@ -569,8 +601,11 @@ self.assertTrue((np.ones((20, DIM)) == se.sv.vectors[:20]).all())
         trainables.prepare_vectors(se.sv, 20, update=False)
-        self.assertEqual((20,DIM), se.sv.vectors.shape)
+        self.assertEqual((20, DIM), se.sv.vectors.shape)
         trainables.prepare_vectors(se.sv, 40, update=True)
-        self.assertEqual((60,DIM), se.sv.vectors.shape)
+        self.assertEqual((60, DIM), se.sv.vectors.shape)
 
-if __name__ == '__main__':
-    logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s', level=logging.DEBUG)
+
+if __name__ == "__main__":
+    logging.basicConfig(
+        format="%(asctime)s : %(levelname)s : %(message)s", level=logging.DEBUG
+    )
     unittest.main()

fse/test/test_inputs.py

 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
 
-# Author: Oliver Borchers <[e-mail]>
-# Copyright (C) 2019 Oliver Borchers
+# Author: Oliver Borchers
+# Copyright (C) Oliver Borchers Oliver Borchers
 
@@ -17,9 +17,17 @@
 
-from fse.inputs import BaseIndexedList, IndexedList, SplitIndexedList, CSplitIndexedList,  \
-    CIndexedList, CSplitCIndexedList, IndexedLineDocument, SplitCIndexedList
+from fse.inputs import (
+    BaseIndexedList,
+    IndexedList,
+    SplitIndexedList,
+    CSplitIndexedList,
+    CIndexedList,
+    CSplitCIndexedList,
+    IndexedLineDocument,
+    SplitCIndexedList,
+)
 
 logger = logging.getLogger(__name__)
 
+
 class TestBaseIndexedList(unittest.TestCase):
-
     def setUp(self):
@@ -33,3 +41,3 @@ self.list_a = ["the dog is good", "it's nice and comfy"]
         self.ll = BaseIndexedList(self.list_a, self.list_b, self.list_c)
-    
+
     def test_init(self):
@@ -56,3 +64,3 @@ _ = BaseIndexedList(self.list_a)
             self.l._check_str_type([])
-    
+
     def test__len(self):
@@ -62,4 +70,3 @@ self.assertEqual(2, len(self.l))
     def test__str(self):
-        self.assertEqual("[\'the dog is good\', \"it\'s nice and comfy\"]",
-        str(self.l))
+        self.assertEqual("['the dog is good', \"it's nice and comfy\"]", str(self.l))
 
@@ -77,7 +84,7 @@ def test__getitem(self):
         self.assertEqual("is it me?", self.ll.items[0])
-    
+
     def test_append(self):
         self.ll.append("is it me?")
         self.assertEqual("is it me?", self.ll.items[-1])
-    
+
     def test_extend(self):
@@ -91,6 +98,7 @@ self.ll.extend(self.list_a)
     def test_extend_ndarr(self):
-        l = BaseIndexedList(np.array([str(i) for i in [1,2,3,4]]))
-        l.extend(np.array([str(i) for i in [1,2,3,4]]))
+        l = BaseIndexedList(np.array([str(i) for i in [1, 2, 3, 4]]))
+        l.extend(np.array([str(i) for i in [1, 2, 3, 4]]))
         self.assertEqual(8, len(l))
 
+
 class TestIndexedList(unittest.TestCase):
@@ -112,7 +120,8 @@ def setUp(self):
 
+
 class TestCIndexedList(unittest.TestCase):
     def setUp(self):
         self.list_a = ["The Dog is good", "it's nice and comfy"]
-        self.il = CIndexedList(self.list_a, custom_index=[1,1])
-    
+        self.il = CIndexedList(self.list_a, custom_index=[1, 1])
+
     def test_cust_index(self):
@@ -130,3 +139,3 @@ self.assertEqual(1, self.il[0][1])
             self.il.__setitem__(0, "the")
-    
+
         with self.assertRaises(NotImplementedError):
@@ -139,2 +148,3 @@ self.il.insert(0, "the")
 
+
 class TestCSplitIndexedList(unittest.TestCase):
@@ -147,11 +157,12 @@ def setUp(self):
         return input.lower().split()
-    
+
     def test_getitem(self):
         self.assertEqual("the dog is good".split(), self.il[0][0])
 
+
 class TestSplitCIndexedList(unittest.TestCase):
     def setUp(self):
         self.list_a = ["The Dog is good", "it's nice and comfy"]
-        self.il = SplitCIndexedList(self.list_a, custom_index=[1,1])
-    
+        self.il = SplitCIndexedList(self.list_a, custom_index=[1, 1])
+
     def test_getitem(self):
@@ -165,3 +176,3 @@ self.assertEqual(("The Dog is good".split(), 1), self.il[0])
             self.il.__setitem__(0, "the")
-    
+
         with self.assertRaises(NotImplementedError):
@@ -174,13 +185,16 @@ self.il.insert(0, "the")
 
+
 class TestCSplitCIndexedList(unittest.TestCase):
     def setUp(self):
         self.list_a = ["The Dog is good", "it's nice and comfy"]
-        self.il = CSplitCIndexedList(self.list_a, custom_split=self.split_func, custom_index=[1,1])
+        self.il = CSplitCIndexedList(
+            self.list_a, custom_split=self.split_func, custom_index=[1, 1]
+        )
 
     def split_func(self, input):
         return input.lower().split()
-    
+
     def test_getitem(self):
         self.assertEqual(("the dog is good".split(), 1), self.il[0])
-    
+
     def test_mutable_funcs(self):
@@ -191,3 +205,3 @@ with self.assertRaises(NotImplementedError):
             self.il.__setitem__(0, "the")
-    
+
         with self.assertRaises(NotImplementedError):
@@ -200,4 +214,4 @@ self.il.insert(0, "the")
 
+
 class TestIndexedLineDocument(unittest.TestCase):
-
     def setUp(self):
@@ -210,3 +224,5 @@ self.p = "fse/test/test_data/test_sentences.txt"
         self.assertEqual("Save yourself money and buy it direct from lg", self.doc[19])
-        self.assertEqual("I am not sure if it is a tracfone problem or the battery", self.doc[-1])
+        self.assertEqual(
+            "I am not sure if it is a tracfone problem or the battery", self.doc[-1]
+        )
 
@@ -223,4 +239,6 @@ def test_yield(self):
 
-if __name__ == '__main__':
-    logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s', level=logging.DEBUG)
-    unittest.main()
+if __name__ == "__main__":
+    logging.basicConfig(
+        format="%(asctime)s : %(levelname)s : %(message)s", level=logging.DEBUG
+    )
+    unittest.main()

fse/test/test_sentencevectors.py

 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
 
-# Author: Oliver Borchers <[e-mail]>
-# Copyright (C) 2019 Oliver Borchers
+# Author: Oliver Borchers
+# Copyright (C) Oliver Borchers Oliver Borchers
 
@@ -34,10 +34,11 @@
 
+
 class TestSentenceVectorsFunctions(unittest.TestCase):
     def setUp(self):
         self.sv = SentenceVectors(2)
-        self.sv.vectors = np.arange(10).reshape(5,2)
+        self.sv.vectors = np.arange(10).reshape(5, 2)
 
     def test_getitem(self):
-        self.assertTrue(([0,1] == self.sv[0]).all())
-        self.assertTrue(([[0,1],[4,5]] == self.sv[[0,2]]).all())
+        self.assertTrue(([0, 1] == self.sv[0]).all())
+        self.assertTrue(([[0, 1], [4, 5]] == self.sv[[0, 2]]).all())
 
@@ -54,6 +55,6 @@ def test_isin(self):
         v1 = self.sv.vectors[0]
-        v1 = v1 / np.sqrt(np.sum(v1**2))
+        v1 = v1 / np.sqrt(np.sum(v1 ** 2))
 
         v2 = self.sv.vectors[1]
-        v2 = v2 / np.sqrt(np.sum(v2**2))
+        v2 = v2 / np.sqrt(np.sum(v2 ** 2))
 
@@ -65,4 +66,4 @@ self.assertTrue(np.allclose(v1, self.sv.vectors_norm[0]))
     def test_get_vector(self):
-        self.assertTrue(([0,1] == self.sv.get_vector(0)).all())
-        self.assertTrue(([2,3] == self.sv.get_vector(1)).all())
+        self.assertTrue(([0, 1] == self.sv.get_vector(0)).all())
+        self.assertTrue(([2, 3] == self.sv.get_vector(1)).all())
 
@@ -98,6 +99,4 @@ def test_init_sims_w_replace(self):
         sv.vectors = np.ones(shape, dtype=np.float32)
-        
-        memvecs = np.memmap(
-            p_target, dtype=np.float32,
-            mode='w+', shape=shape)
+
+        memvecs = np.memmap(p_target, dtype=np.float32, mode="w+", shape=shape)
         memvecs[:] = sv.vectors[:]
@@ -122,9 +121,9 @@ del memvecs
         v1 = self.sv.vectors[0]
-        v1 = v1 / np.sqrt(np.sum(v1**2))
+        v1 = v1 / np.sqrt(np.sum(v1 ** 2))
 
         v2 = self.sv.vectors[1]
-        v2 = v2 / np.sqrt(np.sum(v2**2))
+        v2 = v2 / np.sqrt(np.sum(v2 ** 2))
 
-        self.assertTrue(np.allclose(v1.dot(v2), self.sv.similarity(0,1)))
-        self.assertTrue(np.allclose(1-v1.dot(v2), self.sv.distance(0,1)))
+        self.assertTrue(np.allclose(v1.dot(v2), self.sv.similarity(0, 1)))
+        self.assertTrue(np.allclose(1 - v1.dot(v2), self.sv.distance(0, 1)))
 
@@ -161,3 +160,3 @@ def test_most_similar(self):
         m.sv.init_sims()
-        v = m.sv[[0,1]]
+        v = m.sv[[0, 1]]
         o = m.sv.most_similar(positive=v)
@@ -170,2 +169,3 @@ self.assertEqual(1, o[0][0])
             pass
+
         sentences = IndexedLineDocument(CORPUS)
@@ -203,3 +203,5 @@ m = Average(W2V)
 
-        o = m.sv.most_similar(positive=1, topn=20, restrict_size=(5, 25), indexable=sentences)
+        o = m.sv.most_similar(
+            positive=1, topn=20, restrict_size=(5, 25), indexable=sentences
+        )
         self.assertEqual(20, len(o))
@@ -231,16 +233,27 @@ self.assertEqual(9, o[0][1])
 
+    def test_similar_by_sentence_wrong_model(self):
+        sentences = IndexedLineDocument(CORPUS)
+        m = Average(W2V)
+        m.train(sentences)
+        with self.assertRaises(RuntimeError):
+            m.sv.similar_by_sentence(
+                sentence=["the", "product", "is", "good"], model=W2V
+            )
+
     def test_l2_norm(self):
-        out = np.random.normal(size=(200,50)).astype(np.float32)
+        out = np.random.normal(size=(200, 50)).astype(np.float32)
         result = _l2_norm(out, False)
-        lens = np.sqrt(np.sum((result**2), axis=-1))
+        lens = np.sqrt(np.sum((result ** 2), axis=-1))
         self.assertTrue(np.allclose(1, lens, atol=1e-6))
 
-        out = np.random.normal(size=(200,50)).astype(np.float32)
+        out = np.random.normal(size=(200, 50)).astype(np.float32)
         out = _l2_norm(out, True)
-        lens = np.sqrt(np.sum((out**2), axis=-1))
+        lens = np.sqrt(np.sum((out ** 2), axis=-1))
         self.assertTrue(np.allclose(1, lens, atol=1e-6))
 
 
-if __name__ == '__main__':
-    logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s', level=logging.DEBUG)
+if __name__ == "__main__":
+    logging.basicConfig(
+        format="%(asctime)s : %(levelname)s : %(message)s", level=logging.DEBUG
+    )
     unittest.main()

fse/test/test_sif.py

@@ -8,3 +8,7 @@ import logging
 
-from fse.models.sif import SIF, compute_principal_components, remove_principal_components
+from fse.models.sif import (
+    SIF,
+    compute_principal_components,
+    remove_principal_components,
+)
 from fse.inputs import IndexedLineDocument
@@ -27,6 +31,6 @@
         self.model = SIF(W2V, lang_freq="en")
-    
+
     def test_parameter_sanity(self):
         with self.assertRaises(ValueError):
-            m = SIF(W2V, alpha= -1)
+            m = SIF(W2V, alpha=-1)
             m._check_parameter_sanity()
@@ -48,3 +52,3 @@ with self.assertRaises(ValueError):
         self.assertTrue(np.allclose(self.model.sv.vectors, 0, atol=1e-5))
-    
+
     def test_post_train_calls_no_removal(self):
@@ -55,3 +59,3 @@ self.model.components = 0
         self.assertTrue(np.allclose(self.model.sv.vectors, 1, atol=1e-5))
-    
+
     def test_post_inference_calls(self):
@@ -70,3 +74,3 @@ self.model.sv.vectors = np.ones((200, 10), dtype=np.float32)
             self.model._post_inference_calls(output=None)
-    
+
     def test_post_inference_calls_no_removal(self):
@@ -83,5 +87,8 @@ self.model.components = 0
             self.model._check_dtype_santiy()
-    
+
     def test_dtype_sanity_svd_vals(self):
-        self.model.svd_res = (np.ones_like(self.model.word_weights, dtype=int), np.array(0, dtype=np.float32))
+        self.model.svd_res = (
+            np.ones_like(self.model.word_weights, dtype=int),
+            np.array(0, dtype=np.float32),
+        )
         with self.assertRaises(TypeError):
@@ -91,6 +98,9 @@ self.model._check_dtype_santiy()
     def test_dtype_sanity_svd_vecs(self):
-        self.model.svd_res = (np.array(0, dtype=np.float32), np.ones_like(self.model.word_weights, dtype=int))
+        self.model.svd_res = (
+            np.array(0, dtype=np.float32),
+            np.ones_like(self.model.word_weights, dtype=int),
+        )
         with self.assertRaises(TypeError):
             self.model._check_dtype_santiy()
-    
+
     def test_compute_sif_weights(self):
@@ -109,7 +119,33 @@ cs = 1095661426
         output = self.model.train(self.sentences)
-        self.assertEqual((100,1450), output)
+        self.assertEqual((100, 1450), output)
         self.assertTrue(np.isfinite(self.model.sv.vectors).all())
+        self.assertEqual(2, len(self.model.svd_res))
 
-if __name__ == '__main__':
-    logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s', level=logging.DEBUG)
-    unittest.main()
+    def test_save_issue(self):
+        model = SIF(W2V)
+        model.train(self.sentences)
+
+        p = Path("fse/test/test_data/test_emb.model")
+        model.save(str(p))
+        model = SIF.load(str(p))
+        p.unlink()
+
+        self.assertEqual(2, len(model.svd_res))
+        model.sv.similar_by_sentence("test sentence".split(), model=model)
+
+    def test_broken_vocab(self):
+        w2v = Word2Vec(min_count=1, size=DIM)
+        w2v.build_vocab([l.split() for l in open(CORPUS, "r")])
+        for k in w2v.wv.vocab:
+            w2v.wv.vocab[k].count = np.nan
+
+        model = SIF(w2v)
+        with self.assertRaises(RuntimeError):
+            model.train(self.sentences)
+
+
+if __name__ == "__main__":
+    logging.basicConfig(
+        format="%(asctime)s : %(levelname)s : %(message)s", level=logging.DEBUG
+    )
+    unittest.main()

fse/test/test_usif.py

@@ -21,2 +21,3 @@ import logging
 
+
 class TestuSIFFunctions(unittest.TestCase):
@@ -26,6 +27,6 @@ def setUp(self):
         self.model = uSIF(W2V, lang_freq="en")
-    
+
     def test_parameter_sanity(self):
         with self.assertRaises(ValueError):
-            m = uSIF(W2V, length= 0)
+            m = uSIF(W2V, length=0)
             m._check_parameter_sanity()
@@ -45,3 +46,3 @@ with self.assertRaises(ValueError):
         self.assertTrue(np.allclose(self.model.sv.vectors, 0, atol=1e-5))
-    
+
     def test_post_train_calls_no_removal(self):
@@ -52,3 +53,3 @@ self.model.components = 0
         self.assertTrue(np.allclose(self.model.sv.vectors, 1, atol=1e-5))
-    
+
     def test_post_inference_calls(self):
@@ -67,3 +68,3 @@ self.model.sv.vectors = np.ones((200, 10), dtype=np.float32)
             self.model._post_inference_calls(output=None)
-    
+
     def test_post_inference_calls_no_removal(self):
@@ -80,5 +81,8 @@ self.model.components = 0
             self.model._check_dtype_santiy()
-    
+
     def test_dtype_sanity_svd_vals(self):
-        self.model.svd_res = (np.ones_like(self.model.word_weights, dtype=int), np.array(0, dtype=np.float32))
+        self.model.svd_res = (
+            np.ones_like(self.model.word_weights, dtype=int),
+            np.array(0, dtype=np.float32),
+        )
         with self.assertRaises(TypeError):
@@ -88,6 +92,9 @@ self.model._check_dtype_santiy()
     def test_dtype_sanity_svd_vecs(self):
-        self.model.svd_res = (np.array(0, dtype=np.float32), np.ones_like(self.model.word_weights, dtype=int))
+        self.model.svd_res = (
+            np.array(0, dtype=np.float32),
+            np.ones_like(self.model.word_weights, dtype=int),
+        )
         with self.assertRaises(TypeError):
             self.model._check_dtype_santiy()
-    
+
     def test_compute_usif_weights(self):
@@ -99,3 +106,3 @@ w = "Good"
         a = 0.17831555484795414
-        usif = a / ((a/2) + pw)
+        usif = a / ((a / 2) + pw)
         self.model._compute_usif_weights()
@@ -106,7 +113,21 @@ self.assertTrue(np.allclose(self.model.word_weights[idx], usif))
         output = self.model.train(self.sentences)
-        self.assertEqual((100,1450), output)
+        self.assertEqual((100, 1450), output)
         self.assertTrue(np.isfinite(self.model.sv.vectors).all())
 
-if __name__ == '__main__':
-    logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s', level=logging.DEBUG)
-    unittest.main()
+    def test_broken_vocab(self):
+        w2v = Word2Vec(min_count=1, size=DIM)
+        w2v.build_vocab([l.split() for l in open(CORPUS, "r")])
+        for k in w2v.wv.vocab:
+            w2v.wv.vocab[k].count = np.nan
+
+        model = uSIF(w2v)
+
+        with self.assertRaises(RuntimeError):
+            model.train(self.sentences)
+
+
+if __name__ == "__main__":
+    logging.basicConfig(
+        format="%(asctime)s : %(levelname)s : %(message)s", level=logging.DEBUG
+    )
+    unittest.main()

fse/test/test_utils.py

@@ -5,2 +5,3 @@ import logging
 import numpy as np
+from numpy.testing import assert_allclose, assert_raises
 
@@ -13,6 +14,5 @@ from fse.models.utils import compute_principal_components, remove_principal_components
 class TestUtils(unittest.TestCase):
-
     def test_compute_components(self):
         m = np.random.uniform(size=(500, 10)).astype(np.float32)
-        out = compute_principal_components(vectors = m)
+        out = compute_principal_components(vectors=m)
         self.assertEqual(2, len(out))
@@ -23,27 +23,75 @@ self.assertEqual(1, len(out[1]))
         m = np.random.uniform(size=(500, 10))
-        out = compute_principal_components(vectors = m, components=5)
+        out = compute_principal_components(vectors=m, components=5)
         self.assertEqual(2, len(out))
         self.assertEqual(5, len(out[1]))
-    
+
+    def test_compute_large_components(self):
+        m = np.random.uniform(size=(int(2e6), 100)).astype(np.float32)
+        out = compute_principal_components(vectors=m, cache_size_gb=0.2)
+        self.assertEqual(2, len(out))
+        self.assertEqual(1, len(out[1]))
+        self.assertEqual(np.float32, out[1].dtype)
+
     def test_remove_components_inplace(self):
-        m = np.ones((500,10), dtype=np.float32)
-        out = compute_principal_components(vectors = m)
+        m = np.ones((500, 10), dtype=np.float32)
+        c = np.copy(m)
+        out = compute_principal_components(vectors=m)
         remove_principal_components(m, svd_res=out)
-        self.assertTrue(np.allclose(0., m, atol=1e-5))
-    
+        assert_allclose(m, 0.0, atol=1e-5)
+        with assert_raises(AssertionError):
+            assert_allclose(m, c)
+
     def test_remove_components(self):
-        m = np.ones((500,10), dtype=np.float32)
-        out = compute_principal_components(vectors = m)
+        m = np.ones((500, 10), dtype=np.float32)
+        c = np.copy(m)
+        out = compute_principal_components(vectors=m)
         res = remove_principal_components(m, svd_res=out, inplace=False)
-        self.assertTrue(np.allclose(1., res, atol=1e-5))
+        assert_allclose(res, 0.0, atol=1e-5)
+        assert_allclose(m, c)
 
-    def test_remove_weighted_components(self):
-        m = np.ones((500,10), dtype=np.float32)
-        out = compute_principal_components(vectors = m)
+    def test_remove_weighted_components_inplace(self):
+        m = np.ones((500, 10), dtype=np.float32)
+        c = np.copy(m)
+        out = compute_principal_components(vectors=m)
         remove_principal_components(m, svd_res=out, weights=np.array([0.5]))
-        self.assertTrue(np.allclose(0.75, m))
+        assert_allclose(m, 0.75, atol=1e-5)
+        with assert_raises(AssertionError):
+            assert_allclose(m, c)
 
+    def test_remove_weighted_components(self):
+        m = np.ones((500, 10), dtype=np.float32)
+        c = np.copy(m)
+        out = compute_principal_components(vectors=m)
+        res = remove_principal_components(
+            m, svd_res=out, weights=np.array([0.5]), inplace=False
+        )
+        assert_allclose(res, 0.75, atol=1e-5)
+        assert_allclose(m, c)
 
-if __name__ == '__main__':
-    logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s', level=logging.DEBUG)
-    unittest.main()
+    def test_madvise(self):
+        from pathlib import Path
+        from sys import platform
+        from fse.models.utils import set_madvise_for_mmap
+
+        if platform in ["linux", "linux2", "darwin", "aix"]:
+            p = Path("fse/test/test_data/test_vectors")
+            madvise = set_madvise_for_mmap(True)
+            shape = (500, 10)
+            mat = np.random.normal(size=shape)
+            memvecs = np.memmap(p, dtype=np.float32, mode="w+", shape=shape)
+            memvecs[:] = mat[:]
+            del memvecs
+
+            mat = np.memmap(p, dtype=np.float32, mode="r", shape=shape)
+
+            self.assertEqual(
+                madvise(mat.ctypes.data, mat.size * mat.dtype.itemsize, 1), 0
+            )
+            p.unlink()
+
+
+if __name__ == "__main__":
+    logging.basicConfig(
+        format="%(asctime)s : %(levelname)s : %(message)s", level=logging.DEBUG
+    )
+    unittest.main()

MANIFEST.in

@@ -6,4 +6,3 @@ recursive-include fse/test/test_data *
 
-include fse/models/average_inner.c
 include fse/models/average_inner.pyx
 include fse/models/average_inner.pxd

PKG-INFO

 Metadata-Version: 1.0
 Name: fse
-Version: 0.1.15
+Version: 0.1.17
 Summary: Fast Sentence Embeddings for Gensim
-Home-page: https://github.com/oborchers/Fast_Sentence_Embeddings
+Home-page: UNKNOWN
 Author: Oliver Borchers
-Author-email: [e-mail]
-License: GPL-3.0
+Author-email: [e-mail]
+License: UNKNOWN
 Description: UNKNOWN
 Platform: UNKNOWN

README.md

@@ -1,3 +0,9 @@
-[![Build Status](https://travis-ci.com/oborchers/Fast_Sentence_Embeddings.svg?branch=master)](https://travis-ci.com/oborchers/Fast_Sentence_Embeddings)
-[![Coverage Status](https://coveralls.io/repos/github/oborchers/Fast_Sentence_Embeddings/badge.svg?branch=master)](https://coveralls.io/github/oborchers/Fast_Sentence_Embeddings?branch=master)
+<p align="center">
+<a href="https://travis-ci.com/oborchers/Fast_Sentence_Embeddings"><img alt="Build Status" src="https://travis-ci.com/oborchers/Fast_Sentence_Embeddings.svg?branch=master"></a>
+<a href="https://coveralls.io/github/oborchers/Fast_Sentence_Embeddings?branch=master"><img alt="Coverage Status" src="https://coveralls.io/repos/github/oborchers/Fast_Sentence_Embeddings/badge.svg?branch=master"></a>
+<a href="https://pepy.tech/project/fse"><img alt="Downloads" src="https://pepy.tech/badge/fse"></a>
+<a href="https://lgtm.com/projects/g/oborchers/Fast_Sentence_Embeddings/context:python"><img alt="Language grade: Python" src="https://img.shields.io/lgtm/grade/python/g/oborchers/Fast_Sentence_Embeddings.svg"></a>
+<a href="https://github.com/psf/black"><img alt="Code style: black" src="https://img.shields.io/badge/code%20style-black-000000.svg"></a>
+<a href="https://img.shields.io/github/license/oborchers/Fast_Sentence_Embeddings.svg?style=flat"><img alt="License: GPL3" src="https://img.shields.io/github/license/oborchers/Fast_Sentence_Embeddings.svg?style=flat"></a>
+</p>
 
@@ -9,8 +15,26 @@ Fast Sentence Embeddings (fse)
 
+**Disclaimer**: I am working full time. Unfortunately, I have yet to find time to add all the features I'd like to see. Especially the API needs some overhaul and we need support for gensim 4.0.0.
 
+I am looking for active contributors to keep this package alive. Please feel free to ping me at <[e-mail]> if you are interested.
+
+Audience
+------------
+
+This package builds upon Gensim and is intenteded to compute sentence/paragraph vectors for large databases. Use this package if:
+- (Sentence) Transformers are too slow
+- Your dataset is too large for existing solutions (spacy)
+- Using GPUs is not an option.
+
+The average (online) inference time for a well optimized (and batched) sentence-transformer is around 1ms-10ms per sentence.
+If that is not enough and you are willing to sacrifice a bit in terms of quality, this is your package.
+
+
 Features
 ------------
 
-Find the corresponding blog post(s) here: https://medium.com/@oliverbor/fse-2b1ffa791cf9 (will be updated soon)
+Find the corresponding blog post(s) here (code may be outdated):
 
+- [Visualizing 100,000 Amazon Products](https://towardsdatascience.com/vis-amz-83dea6fcb059)
+- [Sentence Embeddings. Fast, please!](https://towardsdatascience.com/fse-2b1ffa791cf9)
+
 **fse** implements three algorithms for sentence embeddings. You can choose
@@ -51,2 +75,3 @@ between *unweighted sentence averages*,  *smooth inverse frequency averages*, and *unsupervised smooth inverse frequency averages*.
 
+
 Installation
@@ -56,3 +81,3 @@ ------------
 This software depends on NumPy, Scipy, Scikit-learn, Gensim, and Wordfreq. 
-You must have them installed prior to installing fse. Required Python version is 3.6.
+You must have them installed prior to installing fse.
 
@@ -144,2 +169,7 @@ As with gensim, it is also recommended you install a BLAS library before installing fse.
 
+0.1.17:
+- Fixed dependency issue where you cannot install fse properly
+- Updated readme
+- Updated travis python versions (3.6, 3.9)
+
 0.1.15 from 0.1.11:
@@ -174,5 +204,5 @@ - Fixed major FT Ngram computation bug
 
-Author: Oliver Borchers <[e-mail]>
+Author: Oliver Borchers
 
-Copyright (C) 2019 Oliver Borchers
+Copyright (C) 2021 Oliver Borchers
 
@@ -179,0 +209,0 @@ Citation

setup.py

 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
 
-'''
-Run with:
+# Author: Oliver Borchers
+# For License information, see corresponding LICENSE file.
 
-sudo python ./setup.py install
-'''
+"""Template setup.py Read more on
+https://docs.python.org/3.7/distutils/setupscript.html."""
 
+import distutils
+import itertools
 import os
 import platform
-import sys
-import warnings
-from setuptools import setup, find_packages, Extension
+import shutil
+
+from setuptools import Extension, find_packages, setup
 from setuptools.command.build_ext import build_ext
 
-if sys.version_info[:2] < (3, 6):
-    raise Exception('This version of fse needs Python 3.6 or later.')
+NAME = "fse"
+VERSION = "0.1.17"
+DESCRIPTION = "Fast Sentence Embeddings for Gensim"
+AUTHOR = "Oliver Borchers"
+AUTHOR_EMAIL = "[e-mail]"
+URL = "https://github.com/oborchers/Fast_Sentence_Embeddings"
+LICENSE = "GPL-3.0"
+REQUIRES_PYTHON = ">=3.6"
+NUMPY_STR = "numpy >= 1.11.3"
+CYTHON_STR = "Cython==0.29.14"
 
-class custom_build_ext(build_ext):
-    '''Allow C extension building to fail.
-    '''
-    warning_message = '''
-********************************************************************
-WARNING: %s could not be compiled. %s
+INSTALL_REQUIRES = [
+    NUMPY_STR,
+    "scipy >= 0.18.1",
+    "smart_open >= 1.5.0",
+    "scikit-learn >= 0.19.1",
+    "gensim<4",
+    "wordfreq >= 2.2.1",
+    "psutil",
+]
+SETUP_REQUIRES = [NUMPY_STR]
 
-Here are some hints for popular operating systems:
+c_extensions = {
+    "fse.models.average_inner": "fse/models/average_inner.c",
+}
+cpp_extensions = {}
 
-If you are seeing this message on Linux you probably need to
-install GCC and/or the Python development package for your
-version of Python.
 
-Debian and Ubuntu users should issue the following command:
+def need_cython():
+    """Return True if we need Cython to translate any of the extensions.
 
-    $ sudo apt-get install build-essential python-dev
+    If the extensions have already been translated to C/C++, then we don"t need to
+    install Cython and perform the translation.
+    """
+    expected = list(c_extensions.values()) + list(cpp_extensions.values())
+    return any([not os.path.isfile(f) for f in expected])
 
-RedHat, CentOS, and Fedora users should issue the following command:
 
-    $ sudo yum install gcc python-devel
+def make_c_ext(use_cython=False):
+    for module, source in c_extensions.items():
+        if use_cython:
+            source = source.replace(".c", ".pyx")
+        extra_args = []
+        #        extra_args.extend(["-g", "-O0"])  # uncomment if optimization limiting crash info
+        yield Extension(
+            module,
+            sources=[source],
+            language="c",
+            extra_compile_args=extra_args,
+        )
 
-If you are seeing this message on OSX please read the documentation
-here:
 
-http://api.mongodb.org/python/current/installation.html#osx
-********************************************************************
-'''
+def make_cpp_ext(use_cython=False):
+    extra_args = []
+    system = platform.system()
 
-    def run(self):
-        try:
-            build_ext.run(self)
-        except Exception:
-            e = sys.exc_info()[1]
-            sys.stdout.write('%s\n' % str(e))
-            warnings.warn(
-                self.warning_message +
-                'Extension modules' +
-                'There was an issue with your platform configuration - see above.')
+    if system == "Linux":
+        extra_args.append("-std=c++11")
+    elif system == "Darwin":
+        extra_args.extend(["-stdlib=libc++", "-std=c++11"])
+    # extra_args.extend(["-g", "-O0"])  # uncomment if
+    # optimization limiting crash info
+    for module, source in cpp_extensions.items():
+        if use_cython:
+            source = source.replace(".cpp", ".pyx")
+        yield Extension(
+            module,
+            sources=[source],
+            language="c++",
+            extra_compile_args=extra_args,
+            extra_link_args=extra_args,
+        )
 
-    def build_extension(self, ext):
-        name = ext.name
-        try:
-            build_ext.build_extension(self, ext)
-        except Exception:
-            e = sys.exc_info()[1]
-            sys.stdout.write('%s\n' % str(e))
-            warnings.warn(
-                self.warning_message +
-                'The %s extension module' % (name,) +
-                'The output above this warning shows how the compilation failed.')
 
+#
+# We use use_cython=False here for two reasons:
+#
+# 1. Cython may not be available at this stage
+# 2. The actual translation from Cython to C/C++ happens inside CustomBuildExt
+#
+ext_modules = list(
+    itertools.chain(make_c_ext(use_cython=False), make_cpp_ext(use_cython=False))
+)
+
+
+class CustomBuildExt(build_ext):
+    """Custom build_ext action with bootstrapping.
+
+    We need this in order to use numpy and Cython in this script without importing them
+    at module level, because they may not be available yet.
+    """
+
+    #
+    # http://stackoverflow.com/questions/19919905/how-to-bootstrap-numpy-installation-in-setup-py
+    #
     def finalize_options(self):
         build_ext.finalize_options(self)
-        if isinstance(__builtins__, dict):
-            __builtins__['__NUMPY_SETUP__'] = False
-        else:
-            __builtins__.__NUMPY_SETUP__ = False
+        # Prevent numpy from thinking it is still in its setup process:
+        # https://docs.python.org/2/library/__builtin__.html#module-__builtin__
+        __builtins__.__NUMPY_SETUP__ = False
 
         import numpy
+
         self.include_dirs.append(numpy.get_include())
 
-mod_dir = os.path.join(os.path.dirname(__file__), 'fse', 'models')
-fse_dir = os.path.join(os.path.dirname(__file__), 'fse')
+        if need_cython():
+            import Cython.Build
 
-cmdclass = {'build_ext': custom_build_ext}
+            Cython.Build.cythonize(list(make_c_ext(use_cython=True)))
+            Cython.Build.cythonize(list(make_cpp_ext(use_cython=True)))
 
-setup(
-    name='fse',
-    version='0.1.15',
-    description='Fast Sentence Embeddings for Gensim',
 
-    author=u'Oliver Borchers',
-    author_email='[e-mail]',
+class CleanExt(distutils.cmd.Command):
+    description = "Remove C sources, C++ sources and binaries for gensim extensions"
+    user_options = []
 
-    url="https://github.com/oborchers/Fast_Sentence_Embeddings",
+    def initialize_options(self):
+        pass
 
-    license='GPL-3.0',
+    def finalize_options(self):
+        pass
 
-    ext_modules=[
-        Extension('fse.models.average_inner',
-                sources=['./fse/models/average_inner.c'],
-                include_dirs=[mod_dir]),
-        ],
-        
-    cmdclass=cmdclass,
-    packages=find_packages(),
+    def run(self):
+        for root, dirs, files in os.walk("gensim"):
+            files = [
+                os.path.join(root, f)
+                for f in files
+                if os.path.splitext(f)[1] in (".c", ".cpp", ".so")
+            ]
+            for f in files:
+                self.announce("removing %s" % f, level=distutils.log.INFO)
+                os.unlink(f)
 
-    zip_safe=False,
+        if os.path.isdir("build"):
+            self.announce("recursively removing build", level=distutils.log.INFO)
+            shutil.rmtree("build")
 
-    test_suite="fse.test",
 
-    install_requires=[
-        'numpy >= 1.11.3',
-        'scipy >= 0.18.1',
-        'smart_open >= 1.5.0',
-        'scikit-learn >= 0.19.1',
-        'gensim >= 3.8.0',
-        'wordfreq >= 2.2.1',
-        'psutil'
-    ],
+cmdclass = {"build_ext": CustomBuildExt, "clean_ext": CleanExt}
+
+if need_cython():
+    INSTALL_REQUIRES.append(CYTHON_STR)
+    SETUP_REQUIRES.append(CYTHON_STR)
+
+setup(
+    name=NAME,
+    version=VERSION,
+    description=DESCRIPTION,
+    author=AUTHOR,
+    author_email=AUTHOR_EMAIL,
+    packages=find_packages(),
+    requires_python=REQUIRES_PYTHON,
+    install_requires=INSTALL_REQUIRES,
+    setup_requires=SETUP_REQUIRES,
+    ext_modules=ext_modules,
+    cmdclass=cmdclass,
+    zip_safe=False,
     include_package_data=True,
 )

.gitattributes

-# Auto detect text files and perform LF normalization
-* text=auto

.gitignore

-# Compiled source #
-###################
-*.com
-*.class
-*.dll
-*.exe
-*.o
-*.so
-*.pyc
-
-# Packages #
-############
-# it's better to unpack these files and commit the raw source
-# git has its own built in compression methods
-*.7z
-*.dmg
-*.gz
-*.iso
-*.jar
-*.rar
-*.tar
-*.zip
-
-# Logs and databases #
-######################
-*.log
-*.sql
-*.sqlite
-*.pkl
-*.bak
-*.npy
-*.npz
-*.code-workspace
-
-# OS generated files #
-######################
-.DS_Store?
-.DS_Store
-ehthumbs.db
-Icon?
-Thumbs.db
-*.icloud
-
-# Folders #
-###########
-legacy
-latex
-draft
-fse.egg-info/
-
-# Other #
-#########
-.ipynb_checkpoints/
-.settings/
-.vscode/
-.eggs
-fse*.egg-info
-*.pptx
-*.doc
-*.docx
-*.dict
-.coverage
-*.bak
-/build/
-/dist/
-*.prof
-*.lprof
-*.bin
-*.old
-*.model
-*_out.txt
-*.html
-vectors
-*.vectors
-*.joblib

LICENSE

-GNU GENERAL PUBLIC LICENSE
-                       Version 3, 29 June 2007
-
- Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
- Everyone is permitted to copy and distribute verbatim copies
- of this license document, but changing it is not allowed.
-
-                            Preamble
-
-  The GNU General Public License is a free, copyleft license for
-software and other kinds of works.
-
-  The licenses for most software and other practical works are designed
-to take away your freedom to share and change the works.  By contrast,
-the GNU General Public License is intended to guarantee your freedom to
-share and change all versions of a program--to make sure it remains free
-software for all its users.  We, the Free Software Foundation, use the
-GNU General Public License for most of our software; it applies also to
-any other work released this way by its authors.  You can apply it to
-your programs, too.
-
-  When we speak of free software, we are referring to freedom, not
-price.  Our General Public Licenses are designed to make sure that you
-have the freedom to distribute copies of free software (and charge for
-them if you wish), that you receive source code or can get it if you
-want it, that you can change the software or use pieces of it in new
-free programs, and that you know you can do these things.
-
-  To protect your rights, we need to prevent others from denying you
-these rights or asking you to surrender the rights.  Therefore, you have
-certain responsibilities if you distribute copies of the software, or if
-you modify it: responsibilities to respect the freedom of others.
-
-  For example, if you distribute copies of such a program, whether
-gratis or for a fee, you must pass on to the recipients the same
-freedoms that you received.  You must make sure that they, too, receive
-or can get the source code.  And you must show them these terms so they
-know their rights.
-
-  Developers that use the GNU GPL protect your rights with two steps:
-(1) assert copyright on the software, and (2) offer you this License
-giving you legal permission to copy, distribute and/or modify it.
-
-  For the developers' and authors' protection, the GPL clearly explains
-that there is no warranty for this free software.  For both users' and
-authors' sake, the GPL requires that modified versions be marked as
-changed, so that their problems will not be attributed erroneously to
-authors of previous versions.
-
-  Some devices are designed to deny users access to install or run
-modified versions of the software inside them, although the manufacturer
-can do so.  This is fundamentally incompatible with the aim of
-protecting users' freedom to change the software.  The systematic
-pattern of such abuse occurs in the area of products for individuals to
-use, which is precisely where it is most unacceptable.  Therefore, we
-have designed this version of the GPL to prohibit the practice for those
-products.  If such problems arise substantially in other domains, we
-stand ready to extend this provision to those domains in future versions
-of the GPL, as needed to protect the freedom of users.
-
-  Finally, every program is threatened constantly by software patents.
-States should not allow patents to restrict development and use of
-software on general-purpose computers, but in those that do, we wish to
-avoid the special danger that patents applied to a free program could
-make it effectively proprietary.  To prevent this, the GPL assures that
-patents cannot be used to render the program non-free.
-
-  The precise terms and conditions for copying, distribution and
-modification follow.
-
-                       TERMS AND CONDITIONS
-
-  0. Definitions.
-
-  "This License" refers to version 3 of the GNU General Public License.
-
-  "Copyright" also means copyright-like laws that apply to other kinds of
-works, such as semiconductor masks.
-
-  "The Program" refers to any copyrightable work licensed under this
-License.  Each licensee is addressed as "you".  "Licensees" and
-"recipients" may be individuals or organizations.
-
-  To "modify" a work means to copy from or adapt all or part of the work
-in a fashion requiring copyright permission, other than the making of an
-exact copy.  The resulting work is called a "modified version" of the
-earlier work or a work "based on" the earlier work.
-
-  A "covered work" means either the unmodified Program or a work based
-on the Program.
-
-  To "propagate" a work means to do anything with it that, without
-permission, would make you directly or secondarily liable for
-infringement under applicable copyright law, except executing it on a
-computer or modifying a private copy.  Propagation includes copying,
-distribution (with or without modification), making available to the
-public, and in some countries other activities as well.
-
-  To "convey" a work means any kind of propagation that enables other
-parties to make or receive copies.  Mere interaction with a user through
-a computer network, with no transfer of a copy, is not conveying.
-
-  An interactive user interface displays "Appropriate Legal Notices"
-to the extent that it includes a convenient and prominently visible
-feature that (1) displays an appropriate copyright notice, and (2)
-tells the user that there is no warranty for the work (except to the
-extent that warranties are provided), that licensees may convey the
-work under this License, and how to view a copy of this License.  If
-the interface presents a list of user commands or options, such as a
-menu, a prominent item in the list meets this criterion.
-
-  1. Source Code.
-
-  The "source code" for a work means the preferred form of the work
-for making modifications to it.  "Object code" means any non-source
-form of a work.
-
-  A "Standard Interface" means an interface that either is an official
-standard defined by a recognized standards body, or, in the case of
-interfaces specified for a particular programming language, one that
-is widely used among developers working in that language.
-
-  The "System Libraries" of an executable work include anything, other
-than the work as a whole, that (a) is included in the normal form of
-packaging a Major Component, but which is not part of that Major
-Component, and (b) serves only to enable use of the work with that
-Major Component, or to implement a Standard Interface for which an
-implementation is available to the public in source code form.  A
-"Major Component", in this context, means a major essential component
-(kernel, window system, and so on) of the specific operating system
-(if any) on which the executable work runs, or a compiler used to
-produce the work, or an object code interpreter used to run it.
-
-  The "Corresponding Source" for a work in object code form means all
-the source code needed to generate, install, and (for an executable
-work) run the object code and to modify the work, including scripts to
-control those activities.  However, it does not include the work's
-System Libraries, or general-purpose tools or generally available free
-programs which are used unmodified in performing those activities but
-which are not part of the work.  For example, Corresponding Source
-includes interface definition files associated with source files for
-the work, and the source code for shared libraries and dynamically
-linked subprograms that the work is specifically designed to require,
-such as by intimate data communication or control flow between those
-subprograms and other parts of the work.
-
-  The Corresponding Source need not include anything that users
-can regenerate automatically from other parts of the Corresponding
-Source.
-
-  The Corresponding Source for a work in source code form is that
-same work.
-
-  2. Basic Permissions.
-
-  All rights granted under this License are granted for the term of
-copyright on the Program, and are irrevocable provided the stated
-conditions are met.  This License explicitly affirms your unlimited
-permission to run the unmodified Program.  The output from running a
-covered work is covered by this License only if the output, given its
-content, constitutes a covered work.  This License acknowledges your
-rights of fair use or other equivalent, as provided by copyright law.
-
-  You may make, run and propagate covered works that you do not
-convey, without conditions so long as your license otherwise remains
-in force.  You may convey covered works to others for the sole purpose
-of having them make modifications exclusively for you, or provide you
-with facilities for running those works, provided that you comply with
-the terms of this License in conveying all material for which you do
-not control copyright.  Those thus making or running the covered works
-for you must do so exclusively on your behalf, under your direction
-and control, on terms that prohibit them from making any copies of
-your copyrighted material outside their relationship with you.
-
-  Conveying under any other circumstances is permitted solely under
-the conditions stated below.  Sublicensing is not allowed; section 10
-makes it unnecessary.
-
-  3. Protecting Users' Legal Rights From Anti-Circumvention Law.
-
-  No covered work shall be deemed part of an effective technological
-measure under any applicable law fulfilling obligations under article
-11 of the WIPO copyright treaty adopted on 20 December 1996, or
-similar laws prohibiting or restricting circumvention of such
-measures.
-
-  When you convey a covered work, you waive any legal power to forbid
-circumvention of technological measures to the extent such circumvention
-is effected by exercising rights under this License with respect to
-the covered work, and you disclaim any intention to limit operation or
-modification of the work as a means of enforcing, against the work's
-users, your or third parties' legal rights to forbid circumvention of
-technological measures.
-
-  4. Conveying Verbatim Copies.
-
-  You may convey verbatim copies of the Program's source code as you
-receive it, in any medium, provided that you conspicuously and
-appropriately publish on each copy an appropriate copyright notice;
-keep intact all notices stating that this License and any
-non-permissive terms added in accord with section 7 apply to the code;
-keep intact all notices of the absence of any warranty; and give all
-recipients a copy of this License along with the Program.
-
-  You may charge any price or no price for each copy that you convey,
-and you may offer support or warranty protection for a fee.
-
-  5. Conveying Modified Source Versions.
-
-  You may convey a work based on the Program, or the modifications to
-produce it from the Program, in the form of source code under the
-terms of section 4, provided that you also meet all of these conditions:
-
-    a) The work must carry prominent notices stating that you modified
-    it, and giving a relevant date.
-
-    b) The work must carry prominent notices stating that it is
-    released under this License and any conditions added under section
-    7.  This requirement modifies the requirement in section 4 to
-    "keep intact all notices".
-
-    c) You must license the entire work, as a whole, under this
-    License to anyone who comes into possession of a copy.  This
-    License will therefore apply, along with any applicable section 7
-    additional terms, to the whole of the work, and all its parts,
-    regardless of how they are packaged.  This License gives no
-    permission to license the work in any other way, but it does not
-    invalidate such permission if you have separately received it.
-
-    d) If the work has interactive user interfaces, each must display
-    Appropriate Legal Notices; however, if the Program has interactive
-    interfaces that do not display Appropriate Legal Notices, your
-    work need not make them do so.
-
-  A compilation of a covered work with other separate and independent
-works, which are not by their nature extensions of the covered work,
-and which are not combined with it such as to form a larger program,
-in or on a volume of a storage or distribution medium, is called an
-"aggregate" if the compilation and its resulting copyright are not
-used to limit the access or legal rights of the compilation's users
-beyond what the individual works permit.  Inclusion of a covered work
-in an aggregate does not cause this License to apply to the other
-parts of the aggregate.
-
-  6. Conveying Non-Source Forms.
-
-  You may convey a covered work in object code form under the terms
-of sections 4 and 5, provided that you also convey the
-machine-readable Corresponding Source under the terms of this License,
-in one of these ways:
-
-    a) Convey the object code in, or embodied in, a physical product
-    (including a physical distribution medium), accompanied by the
-    Corresponding Source fixed on a durable physical medium
-    customarily used for software interchange.
-
-    b) Convey the object code in, or embodied in, a physical product
-    (including a physical distribution medium), accompanied by a
-    written offer, valid for at least three years and valid for as
-    long as you offer spare parts or customer support for that product
-    model, to give anyone who possesses the object code either (1) a
-    copy of the Corresponding Source for all the software in the
-    product that is covered by this License, on a durable physical
-    medium customarily used for software interchange, for a price no
-    more than your reasonable cost of physically performing this
-    conveying of source, or (2) access to copy the
-    Corresponding Source from a network server at no charge.
-
-    c) Convey individual copies of the object code with a copy of the
-    written offer to provide the Corresponding Source.  This
-    alternative is allowed only occasionally and noncommercially, and
-    only if you received the object code with such an offer, in accord
-    with subsection 6b.
-
-    d) Convey the object code by offering access from a designated
-    place (gratis or for a charge), and offer equivalent access to the
-    Corresponding Source in the same way through the same place at no
-    further charge.  You need not require recipients to copy the
-    Corresponding Source along with the object code.  If the place to
-    copy the object code is a network server, the Corresponding Source
-    may be on a different server (operated by you or a third party)
-    that supports equivalent copying facilities, provided you maintain
-    clear directions next to the object code saying where to find the
-    Corresponding Source.  Regardless of what server hosts the
-    Corresponding Source, you remain obligated to ensure that it is
-    available for as long as needed to satisfy these requirements.
-
-    e) Convey the object code using peer-to-peer transmission, provided
-    you inform other peers where the object code and Corresponding
-    Source of the work are being offered to the general public at no
-    charge under subsection 6d.
-
-  A separable portion of the object code, whose source code is excluded
-from the Corresponding Source as a System Library, need not be
-included in conveying the object code work.
-
-  A "User Product" is either (1) a "consumer product", which means any
-tangible personal property which is normally used for personal, family,
-or household purposes, or (2) anything designed or sold for incorporation
-into a dwelling.  In determining whether a product is a consumer product,
-doubtful cases shall be resolved in favor of coverage.  For a particular
-product received by a particular user, "normally used" refers to a
-typical or common use of that class of product, regardless of the status
-of the particular user or of the way in which the particular user
-actually uses, or expects or is expected to use, the product.  A product
-is a consumer product regardless of whether the product has substantial
-commercial, industrial or non-consumer uses, unless such uses represent
-the only significant mode of use of the product.
-
-  "Installation Information" for a User Product means any methods,
-procedures, authorization keys, or other information required to install
-and execute modified versions of a covered work in that User Product from
-a modified version of its Corresponding Source.  The information must
-suffice to ensure that the continued functioning of the modified object
-code is in no case prevented or interfered with solely because
-modification has been made.
-
-  If you convey an object code work under this section in, or with, or
-specifically for use in, a User Product, and the conveying occurs as
-part of a transaction in which the right of possession and use of the
-User Product is transferred to the recipient in perpetuity or for a
-fixed term (regardless of how the transaction is characterized), the
-Corresponding Source conveyed under this section must be accompanied
-by the Installation Information.  But this requirement does not apply
-if neither you nor any third party retains the ability to install
-modified object code on the User Product (for example, the work has
-been installed in ROM).
-
-  The requirement to provide Installation Information does not include a
-requirement to continue to provide support service, warranty, or updates
-for a work that has been modified or installed by the recipient, or for
-the User Product in which it has been modified or installed.  Access to a
-network may be denied when the modification itself materially and
-adversely affects the operation of the network or violates the rules and
-protocols for communication across the network.
-
-  Corresponding Source conveyed, and Installation Information provided,
-in accord with this section must be in a format that is publicly
-documented (and with an implementation available to the public in
-source code form), and must require no special password or key for
-unpacking, reading or copying.
-
-  7. Additional Terms.
-
-  "Additional permissions" are terms that supplement the terms of this
-License by making exceptions from one or more of its conditions.
-Additional permissions that are applicable to the entire Program shall
-be treated as though they were included in this License, to the extent
-that they are valid under applicable law.  If additional permissions
-apply only to part of the Program, that part may be used separately
-under those permissions, but the entire Program remains governed by
-this License without regard to the additional permissions.
-
-  When you convey a copy of a covered work, you may at your option
-remove any additional permissions from that copy, or from any part of
-it.  (Additional permissions may be written to require their own
-removal in certain cases when you modify the work.)  You may place
-additional permissions on material, added by you to a covered work,
-for which you have or can give appropriate copyright permission.
-
-  Notwithstanding any other provision of this License, for material you
-add to a covered work, you may (if authorized by the copyright holders of
-that material) supplement the terms of this License with terms:
-
-    a) Disclaiming warranty or limiting liability differently from the
-    terms of sections 15 and 16 of this License; or
-
-    b) Requiring preservation of specified reasonable legal notices or
-    author attributions in that material or in the Appropriate Legal
-    Notices displayed by works containing it; or
-
-    c) Prohibiting misrepresentation of the origin of that material, or
-    requiring that modified versions of such material be marked in
-    reasonable ways as different from the original version; or
-
-    d) Limiting the use for publicity purposes of names of licensors or
-    authors of the material; or
-
-    e) Declining to grant rights under trademark law for use of some
-    trade names, trademarks, or service marks; or
-
-    f) Requiring indemnification of licensors and authors of that
-    material by anyone who conveys the material (or modified versions of
-    it) with contractual assumptions of liability to the recipient, for
-    any liability that these contractual assumptions directly impose on
-    those licensors and authors.
-
-  All other non-permissive additional terms are considered "further
-restrictions" within the meaning of section 10.  If the Program as you
-received it, or any part of it, contains a notice stating that it is
-governed by this License along with a term that is a further
-restriction, you may remove that term.  If a license document contains
-a further restriction but permits relicensing or conveying under this
-License, you may add to a covered work material governed by the terms
-of that license document, provided that the further restriction does
-not survive such relicensing or conveying.
-
-  If you add terms to a covered work in accord with this section, you
-must place, in the relevant source files, a statement of the
-additional terms that apply to those files, or a notice indicating
-where to find the applicable terms.
-
-  Additional terms, permissive or non-permissive, may be stated in the
-form of a separately written license, or stated as exceptions;
-the above requirements apply either way.
-
-  8. Termination.
-
-  You may not propagate or modify a covered work except as expressly
-provided under this License.  Any attempt otherwise to propagate or
-modify it is void, and will automatically terminate your rights under
-this License (including any patent licenses granted under the third
-paragraph of section 11).
-
-  However, if you cease all violation of this License, then your
-license from a particular copyright holder is reinstated (a)
-provisionally, unless and until the copyright holder explicitly and
-finally terminates your license, and (b) permanently, if the copyright
-holder fails to notify you of the violation by some reasonable means
-prior to 60 days after the cessation.
-
-  Moreover, your license from a particular copyright holder is
-reinstated permanently if the copyright holder notifies you of the
-violation by some reasonable means, this is the first time you have
-received notice of violation of this License (for any work) from that
-copyright holder, and you cure the violation prior to 30 days after
-your receipt of the notice.
-
-  Termination of your rights under this section does not terminate the
-licenses of parties who have received copies or rights from you under
-this License.  If your rights have been terminated and not permanently
-reinstated, you do not qualify to receive new licenses for the same
-material under section 10.
-
-  9. Acceptance Not Required for Having Copies.
-
-  You are not required to accept this License in order to receive or
-run a copy of the Program.  Ancillary propagation of a covered work
-occurring solely as a consequence of using peer-to-peer transmission
-to receive a copy likewise does not require acceptance.  However,
-nothing other than this License grants you permission to propagate or
-modify any covered work.  These actions infringe copyright if you do
-not accept this License.  Therefore, by modifying or propagating a
-covered work, you indicate your acceptance of this License to do so.
-
-  10. Automatic Licensing of Downstream Recipients.
-
-  Each time you convey a covered work, the recipient automatically
-receives a license from the original licensors, to run, modify and
-propagate that work, subject to this License.  You are not responsible
-for enforcing compliance by third parties with this License.
-
-  An "entity transaction" is a transaction transferring control of an
-organization, or substantially all assets of one, or subdividing an
-organization, or merging organizations.  If propagation of a covered
-work results from an entity transaction, each party to that
-transaction who receives a copy of the work also receives whatever
-licenses to the work the party's predecessor in interest had or could
-give under the previous paragraph, plus a right to possession of the
-Corresponding Source of the work from the predecessor in interest, if
-the predecessor has it or can get it with reasonable efforts.
-
-  You may not impose any further restrictions on the exercise of the
-rights granted or affirmed under this License.  For example, you may
-not impose a license fee, royalty, or other charge for exercise of
-rights granted under this License, and you may not initiate litigation
-(including a cross-claim or counterclaim in a lawsuit) alleging that
-any patent claim is infringed by making, using, selling, offering for
-sale, or importing the Program or any portion of it.
-
-  11. Patents.
-
-  A "contributor" is a copyright holder who authorizes use under this
-License of the Program or a work on which the Program is based.  The
-work thus licensed is called the contributor's "contributor version".
-
-  A contributor's "essential patent claims" are all patent claims
-owned or controlled by the contributor, whether already acquired or
-hereafter acquired, that would be infringed by some manner, permitted
-by this License, of making, using, or selling its contributor version,
-but do not include claims that would be infringed only as a
-consequence of further modification of the contributor version.  For
-purposes of this definition, "control" includes the right to grant
-patent sublicenses in a manner consistent with the requirements of
-this License.
-
-  Each contributor grants you a non-exclusive, worldwide, royalty-free
-patent license under the contributor's essential patent claims, to
-make, use, sell, offer for sale, import and otherwise run, modify and
-propagate the contents of its contributor version.
-
-  In the following three paragraphs, a "patent license" is any express
-agreement or commitment, however denominated, not to enforce a patent
-(such as an express permission to practice a patent or covenant not to
-sue for patent infringement).  To "grant" such a patent license to a
-party means to make such an agreement or commitment not to enforce a
-patent against the party.
-
-  If you convey a covered work, knowingly relying on a patent license,
-and the Corresponding Source of the work is not available for anyone
-to copy, free of charge and under the terms of this License, through a
-publicly available network server or other readily accessible means,
-then you must either (1) cause the Corresponding Source to be so
-available, or (2) arrange to deprive yourself of the benefit of the
-patent license for this particular work, or (3) arrange, in a manner
-consistent with the requirements of this License, to extend the patent
-license to downstream recipients.  "Knowingly relying" means you have
-actual knowledge that, but for the patent license, your conveying the
-covered work in a country, or your recipient's use of the covered work
-in a country, would infringe one or more identifiable patents in that
-country that you have reason to believe are valid.
-
-  If, pursuant to or in connection with a single transaction or
-arrangement, you convey, or propagate by procuring conveyance of, a
-covered work, and grant a patent license to some of the parties
-receiving the covered work authorizing them to use, propagate, modify
-or convey a specific copy of the covered work, then the patent license
-you grant is automatically extended to all recipients of the covered
-work and works based on it.
-
-  A patent license is "discriminatory" if it does not include within
-the scope of its coverage, prohibits the exercise of, or is
-conditioned on the non-exercise of one or more of the rights that are
-specifically granted under this License.  You may not convey a covered
-work if you are a party to an arrangement with a third party that is
-in the business of distributing software, under which you make payment
-to the third party based on the extent of your activity of conveying
-the work, and under which the third party grants, to any of the
-parties who would receive the covered work from you, a discriminatory
-patent license (a) in connection with copies of the covered work
-conveyed by you (or copies made from those copies), or (b) primarily
-for and in connection with specific products or compilations that
-contain the covered work, unless you entered into that arrangement,
-or that patent license was granted, prior to 28 March 2007.
-
-  Nothing in this License shall be construed as excluding or limiting
-any implied license or other defenses to infringement that may
-otherwise be available to you under applicable patent law.
-
-  12. No Surrender of Others' Freedom.
-
-  If conditions are imposed on you (whether by court order, agreement or
-otherwise) that contradict the conditions of this License, they do not
-excuse you from the conditions of this License.  If you cannot convey a
-covered work so as to satisfy simultaneously your obligations under this
-License and any other pertinent obligations, then as a consequence you may
-not convey it at all.  For example, if you agree to terms that obligate you
-to collect a royalty for further conveying from those to whom you convey
-the Program, the only way you could satisfy both those terms and this
-License would be to refrain entirely from conveying the Program.
-
-  13. Use with the GNU Affero General Public License.
-
-  Notwithstanding any other provision of this License, you have
-permission to link or combine any covered work with a work licensed
-under version 3 of the GNU Affero General Public License into a single
-combined work, and to convey the resulting work.  The terms of this
-License will continue to apply to the part which is the covered work,
-but the special requirements of the GNU Affero General Public License,
-section 13, concerning interaction through a network will apply to the
-combination as such.
-
-  14. Revised Versions of this License.
-
-  The Free Software Foundation may publish revised and/or new versions of
-the GNU General Public License from time to time.  Such new versions will
-be similar in spirit to the present version, but may differ in detail to
-address new problems or concerns.
-
-  Each version is given a distinguishing version number.  If the
-Program specifies that a certain numbered version of the GNU General
-Public License "or any later version" applies to it, you have the
-option of following the terms and conditions either of that numbered
-version or of any later version published by the Free Software
-Foundation.  If the Program does not specify a version number of the
-GNU General Public License, you may choose any version ever published
-by the Free Software Foundation.
-
-  If the Program specifies that a proxy can decide which future
-versions of the GNU General Public License can be used, that proxy's
-public statement of acceptance of a version permanently authorizes you
-to choose that version for the Program.
-
-  Later license versions may give you additional or different
-permissions.  However, no additional obligations are imposed on any
-author or copyright holder as a result of your choosing to follow a
-later version.
-
-  15. Disclaimer of Warranty.
-
-  THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
-APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
-HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
-OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
-THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
-PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
-IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
-ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
-
-  16. Limitation of Liability.
-
-  IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
-WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
-THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
-GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
-USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
-DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
-PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
-EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
-SUCH DAMAGES.
-
-  17. Interpretation of Sections 15 and 16.
-
-  If the disclaimer of warranty and limitation of liability provided
-above cannot be given local legal effect according to their terms,
-reviewing courts shall apply local law that most closely approximates
-an absolute waiver of all civil liability in connection with the
-Program, unless a warranty or assumption of liability accompanies a
-copy of the Program in return for a fee.
-
-                     END OF TERMS AND CONDITIONS
-
-            How to Apply These Terms to Your New Programs
-
-  If you develop a new program, and you want it to be of the greatest
-possible use to the public, the best way to achieve this is to make it
-free software which everyone can redistribute and change under these terms.
-
-  To do so, attach the following notices to the program.  It is safest
-to attach them to the start of each source file to most effectively
-state the exclusion of warranty; and each file should have at least
-the "copyright" line and a pointer to where the full notice is found.
-
-    <one line to give the program's name and a brief idea of what it does.>
-    Copyright (C) <year>  <name of author>
-
-    This program is free software: you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation, either version 3 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License
-    along with this program.  If not, see <https://www.gnu.org/licenses/>.
-
-Also add information on how to contact you by electronic and paper mail.
-
-  If the program does terminal interaction, make it output a short
-notice like this when it starts in an interactive mode:
-
-    <program>  Copyright (C) <year>  <name of author>
-    This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
-    This is free software, and you are welcome to redistribute it
-    under certain conditions; type `show c' for details.
-
-The hypothetical commands `show w' and `show c' should show the appropriate
-parts of the General Public License.  Of course, your program's commands
-might be different; for a GUI interface, you would use an "about box".
-
-  You should also get your employer (if you work as a programmer) or school,
-if any, to sign a "copyright disclaimer" for the program, if necessary.
-For more information on this, and how to apply and follow the GNU GPL, see
-<https://www.gnu.org/licenses/>.
-
-  The GNU General Public License does not permit incorporating your program
-into proprietary programs.  If your program is a subroutine library, you
-may consider it more useful to permit linking proprietary applications with
-the library.  If this is what you want to do, use the GNU Lesser General
-Public License instead of this License.  But first, please read
-<https://www.gnu.org/licenses/why-not-lgpl.html>.

Alert delta unavailable

Currently unable to show alert delta for PyPI packages.

Improved metrics

Lines of code

3737

21.65%

Worsened metrics

Total package byte prevSize

562455

-15.32%

Number of package files

33

-10.81%