pbu

Python Basic Utilities pbu

Available on PyPi

Table of Contents

1. Installation
2. Usage
3. Classes
   1. JSON - a JavaScript-like dictionary access helper
   2. Logger - a wrapper around the Python logging framework
   3. TimeSeries - powerful helper class to organise time series
   4. BasicMonitor - monitor class orchestrating regular operations
   5. ConstantListing - a parent class allowing to fetch attribute values from a constant class
   6. PerformanceLogger - a utility class to log runtime performance of processes
   7. PerformanceTracker - a utility class to track performance of a repeated process
   8. BasicConfig - application utility class managing access to environment variables
   9. JsonDocument - a class that can serialise/deserialise a dictionary into a class instance
4. Functions
   1. list_to_json
   2. json_to_list
   3. default_options
   4. default_value
   5. list_find_one
   6. list_map_filter
   7. list_join
   8. not_none
   9. Datetime Functions
   10. weighted_mean
   11. normalise

Installation

Install via pip:

pip install pbu

Usage

Optional: If you have a requirement.txt file, you can add pbu:

pbu

Then, simply import the class / module you need:

from pbu import JSON

# and start using it
obj = JSON({"my": {"obj": "content"}})
print(obj.my.obj)

Classes

JSON

This is an adaptation of the native dict class, providing Javascript-like dictionary access using the "dot-notation" (e.g. person.relations[0].address.street) rather than the Python-native bracket notation (e.g. person["relations"][0]["address"]["street"]). It overrides the basic __getattr__ and __setattr__ methods as a shortcut to manage the dictionary content.

Example

from pbu import JSON

my_obj = JSON({"initial": "content"})
print(my_obj.initial)
# prints out "content"

my_obj.initial = {"a": 5, "b": 3}
print(my_obj.initial.a + my_obj.initial.b)
# prints out 8
my_obj.initial.b = 13
print(my_obj.initial.a + my_obj.initial.b)
# prints out 18

my_obj.extension = 10
print(my_obj.extension)
# prints out 10

Logger

This is a basic logger allowing to write log files, for logger.info it writes a debug.log and for logger.error or logger.exception it writes an error.log file.

Example

from pbu import Logger

logger = Logger(name="logger-name")
logger.debug("Some debug message goes here")
logger.error("Error executing something")

logger = Logger(name="logger-name", log_folder="./logs")
logger.debug("This will create the debug.log and error.log in the ./logs folder")

TimeSeries

The time series class is a helper utility, that allows to compile complex time-series, offering functionality to add time series, remove time series and most importantly align time series with timestamps to a previously defined resolution by interpolating missing values and re-aligning measurements within the tolerance of the provided time series.

It supports 2 different structures:

List of Dictionary Items

from datetime import datetime, timedelta

list_of_dict = [
    {"date_time": datetime.now(), "measurement_1": 12, "measurement_2": 15},
    {"date_time": datetime.now() + timedelta(hours=1), "measurement_1": 10, "measurement_2": 16},
    {"date_time": datetime.now() + timedelta(hours=2), "measurement_1": 9, "measurement_2": 12},
]

Dictionary of Lists

from datetime import datetime, timedelta

dict_of_list = {
    "date_time": [datetime.now(), datetime.now() + timedelta(hours=1), datetime + timedelta(hours=2)],
    "measurement_1": [12, 10, 16],
    "measurement_2": [15, 16, 12],
}

Example

from pbu import TimeSeries
from datetime import datetime, timedelta

# initial time series base data (you can add measurements as well or provide as list of dictionaries
dict_of_list = {
    "date_time": TimeSeries.create_date_range(datetime.now(), datetime.now() + timedelta(days=1), timedelta(hours=3)),
}

# init time series
ts = TimeSeries(input_data=dict_of_list, date_time_key="date_time")
# add values (ensure same length as date_time series)
ts.add_values("measurement_1", [12, 10, 16, 10, 5, 8, 12, 9])

# you can translate into a list of dictionary items (keys are maintained)
list_of_dict = ts.translate_to_list_of_dicts()

# extract data series from the time series
measurement_1 = ts.get_values("measurement_1")

# create new series that provides same value for all timestamps
ts.fill_values("constant_series", 5)

# remove a series from the total data structure
ts.remove_series("constant_series")

# re-sample data to 5 minute resolution, interpolating values, also pre-pending another day in front of the time series 
ts.align_to_resolution(resolution=timedelta(minutes=5), start_date=datetime.now() - timedelta(days=1))
# this will result in "interpolated" values for the first day, using the first value (12) to fill missing values
print(len(ts.translate_to_list_of_dicts()))  # 12 an hour, 2 days, 48 * 12 = ~576 items

# the same can also be achieved by:
ts.set_resolution(timedelta(minutes=5))
# no need to provide resolution now
ts.align_to_resolution(start_date=datetime.now() - timedelta(days=1))

BasicMonitor

An abstract class providing base-functionality for running monitors - threads that run a specific routine in a regular interval. This can be an executor waiting for new tasks to be processed (and checking every 5 seconds) or a thread that monitors some readout in a regular interval. The monitor is wrapped to re-start itself, in case of errors.

Example

from pbu import BasicMonitor


class MyOwnMonitor(BasicMonitor):
    def __init__(self, data):
        super().__init__(monitor_id="my_id", wait_time=5)  # waits 5 seconds between each execution loop
        self.data = data

    def running(self):
        while self.active:
            # your code goes here (example):
            # result = fetch_data(self.data)
            # store_result(result)
            self.wait()

If you want to run in a regular interval, the running method needs to be slightly modified:

from time import time
from pbu import BasicMonitor


class MyRegularOwnMonitor(BasicMonitor):
    def __init__(self, data):
        super().__init__(monitor_id="another_id", wait_time=60, run_interval=True)  # execute every 60 seconds
        self.data = data

    def running(self):
        while self.active:
            start_ts = time()  # capture start of loop
            # your code goes here (example):
            # result = do_something(self.data)
            # store_result(result)
            self.wait(exec_duration=round(time() - start_ts))  # include the execution duration
            if self.is_interrupted:
                # wait time got interrupted from the outside (see below), flag will be reset on next call to wait()
                pass

Optional constructor parameters

• You can also pass a custom logger as custom_logger argument to the constructor. By default it will use the pbu.Logger and log major events such as start/stop/restart and errors.
• Passing a ping_interval parameter allows you to check for overdue jobs more often than the wait time. For example you could have a wait_time of 1800s (30 min) and a ping_interval of 60s, which allows you to not miss out on an execution if your machine running the monitor should sleep (e.g. on a laptop when you put it on standby, the sleep timer stops). By default this is 60 seconds (or the wait_time, if the wait_time is lower than 60s)

Manage and run monitor

import threading


def start_monitor_thread(monitor):
    """
    Thread function to be run by the new thread.
    :param monitor: BasicMonitor - an instance of sub-class of BasicMonitor 
    """
    # start the monitor
    monitor.start()


# create monitor instance of your own class that implements BasicMonitor
regular_monitor = MyRegularOwnMonitor(data={"some": "data"})

# create thread with start-up function and start it
t = threading.Thread(target=start_monitor_thread, args=(regular_monitor,), daemon=True)
t.start()

# if you want to interrupt the wait time at any point (temporarily sets the .is_interrupted attribute to True)
regular_monitor.interrupt()

# in a separate piece of code (e.g. REST handler or timer) you can stop the monitor instance
regular_monitor.stop()

Stopping a monitor doesn't interrupt the current thread. If the monitor is for example in a wait period and you send the stop signal, the thread will still run until the wait period passes.

In an API scenario, I recommend using a dict or list to cache monitors and retrieve them via the API using the to_json() method for identification. This then allows you to signal starting / stopping of monitors by providing the monitor ID and lookup the monitor instance in the monitor cache.

BasicMonitor Methods

• start() - starts the monitor
• stop() - stops the monitor
• to_json() - returns a dictionary with basic monitor technical information (id, state, wait behaviour, etc)
• wait_till_midnight() - waits till the next midnight in your machines time zone
• wait(exec_duration=0) - waits for the time specified in the constructor and in case of run_interval=True for the optional exec_duration, if provided.

ConstantListing

Managing constants is good practice for avoiding typos. Imagine the following class:

class Tags:
    GEO = "GEO"
    EQUIPMENT = "EQUIPMENT"

This allows you to just do: Tags.GEO allowing you to use your IDEs auto-complete, avoiding typos. But if you want to programmatically get all possible values for Tags, you can use pbu's ConstantListing class:

from pbu import ConstantListing


class Tags(ConstantListing):
    GEO = "GEO"
    EQUIPMENT = "EQUIPMENT"


list_of_values = Tags().get_all()  # will return ['GEO', 'EQUIPMENT']

PerformanceLogger

This utility class allows to print out or log runtime performance expressed as time delta between a start time and an end time.

Basic usage:

from pbu import PerformanceLogger

perf = PerformanceLogger()
perf.start()  # this is optional and will reset the start-time
# do something useful...
perf.checkpoint(message="Step 1")  # will print "Step 1 took <timedelta>
# some some more useful stuff...
perf.finish(message="Something useful")  # will print out the whole duration from start to finish

You can omit the message of a checkpoint call if you don't need an output for an operation, but want to print out the duration of the step that follows.

You can also use a Python Logger object (or pbu.Logger) instead of the message being printed out onto the console.

from pbu import Logger, PerformanceLogger

logger = Logger("my-logger-name")
perf = PerformanceLogger()
# do something...
perf.checkpoint()  # next output will print the duration between this point and the next checkpoint call
# do some more stuff...
perf.checkpoint(message="Some More Stuff", logger=logger)
# and even more ...
perf.finish(message="Total operation", logger=logger)

Methods

• start() - will reset the start time of the performance logger
• checkpoint(message=None, logger=None) - creates a new checkpoint and optionally logs a message
• finish(message=None, logger=None) - prints out the total runtime since start() was called or the class was initialised

PerformanceTracker

A utility class that allows to track the runtime of a repeated process and print out performance stats every n repetitions.

Basic usage:

from pbu import PerformanceTracker

tracker = PerformanceTracker(operation_name="compute", print_interval=20)
for i in range(0, 100):
    # starting the operation is thread-safe and can be executed in parallel, unique keys are getting returned 
    track_key = tracker.start_operation()
    # perform your operation
    a = i * i * i
    tracker.end_operation(track_key)

Every 20 executions, this will print out a line line this:

Performance for operation 'compute' (20): Avg: 5.960464477539062e-07s | Min: 2.384185791015625e-07 | Max: 1.1920928955078125e-06

with the operation name, followed by the number of executions and then avg, min and max performance in seconds.

BasicConfig

This class can be used in applications to simplify access to environment variables. It is recommended to write your own sub-class of this class, where you can provide even more convenient access. However, the class can also be used standalone.

Basic usage:

import os
from pbu import BasicConfig


class Config(BasicConfig):
    def __init__(self):
        super().__init__(default_values={
            "PORT": 5000,
            "IS_DEBUG": 1,
            "DATA_DIRECTORY": None,
        }, directory_keys=["DATA_DIRECTORY"], required=["DATA_DIRECTORY"])

    def get_port(self) -> int:
        return int(self.get_config_value("port"))

    def is_debug(self) -> bool:
        return int(self.get_config_value("is_debug")) == 1

    def get_data_directory(self) -> str:
        return self.get_config_value("DATA_DIRECTORY")


cfg = Config()
# BasicConfig will ensure the directory exists
result = os.path.exists(cfg.get_data_directory())
# result is True      

Methods

• get_config_value(config_key, default_value=None) - retrieves a config value, the default value override is optional as it should already be provided in the default_values of the constructor. If a config_key hasn't been provided by the default_values of the constructor, this will trigger reading the value fresh from the environment and storing it within this class.
• __init__(default_values={}, directory_keys=[], required=[], env_file=".env") - super constructor, which will be used to load the initial environment.
  • The default_values provide the keys that will be extracted from the OS environment.
  • The directory_keys are config keys that will be used to run a directory check. If the provided environment value refers to a directory that doesn't exist yet, the class will attempt to create it.
  • The required parameter provides environment keys that have to be provided by the OS environment. If they are not available in the environment, an EnvironmentError will be raised.

JsonDocument

Methods

• to_json() - call this to return a dict representation of the instance. This will serialise the id and data_model_version attributes and any attributes provided in the get_attribute_mapping() method.
• get_attribute_mapping() - provides a dict mapping between class attributes and JSON keys that will be used in the dict representation.
• extract_system_fields(json: dict) - this will deserialise a dict and map the _id field to the id attribute, dataModelVersion field to data_model_version attribute and any field defined in the get_attribute_mapping() method.
• apply_updates(update, attributes = []) - overwrites attributes of the current instance with the update. The list of attributes has to be specified and is empty by default. The update must be of the same type as the current instance. If an attribute is listed that does not exist, a warning will be issued.

Static Methods

• .from_json(json) - this method has to be implemented by any sub-class and is responsible for deserialising a JSON document into an instance of your sub-class. The instance method extract_system_fields(json) can be used to map most simple attributes - i.e. any attributes provided in the get_attribute_mapping() method.

Functions

list_to_json

from pbu import list_to_json

# assuming we have `my_store` as an instance of MongoDB store or MySQL store, you can:
list_of_dictionaries = list_to_json(item_list=my_store.get_all())  # output is a list of dictionaries

This function operates on lists of objects inheriting from JsonDocument and converts them into dictionaries using the to_json() method of any object passed into the function. Objects passed into the function require the to_json() method and need to return the dictionary representation of the object. This function is just a mapping shortcut.

list_from_json

from pbu import list_from_json

# assuming we have a class `MyClass` that inherits from `JsonDocument` and implements the `from_json()` method
list_from_json(item_list=[{"a": 1, "b": 2}, {"a": 3, "b": 4}], class_type=MyClass)

default_options

from pbu import default_options

DEFAULTS = {
    "a": 1,
    "b": 2,
    "c": 3,
}

result = default_options(default=DEFAULTS, override={"b": 4, "d": 5})
# result is: {"a": 1, "b": 4, "c": 3, "d": 5}

If you want to avoid additional keys other than the keys in DEFAULTS, you can provide a third argument:

from pbu import default_options

DEFAULTS = {
    "a": 1,
    "b": 2,
}

result = default_options(default=DEFAULTS, override={"b": 4, "d": 5}, allow_unknown_keys=False)
# result is: {"a": 1, "b": 4}

default_value

from pbu import default_value

result = default_value(value=None, fallback=5)  # None is by default disallowed
# result is 5

result = default_value(value=0, fallback=5, disallowed=[None, 0])  # either 0 or None would return the fallback
# result is 5

result = default_value(0, 5)  # value will be used, as it doesn't match None
# result is 0

list_find_one

Finds the first item in a list that matches the filter function - this is a shortcut for running filter(..) on a list, then checking its length and if the length is > 0 fetching the first item.

from pbu import list_find_one

my_list = ["a", "b", "c"]

result = list_find_one(lambda x: x == "c", my_list)
# result is "c"

result = list_find_one(lambda x: x == "d", my_list)
# result is None

list_map_filter

A shorthand for filtering and mapping a lsit of items. The function allows to pass both lambdas (filter and map) into one function call. A boolean flag (filter_first=True) decides whether the filter or map operation is called first.

from pbu import list_map_filter

my_list = [
    {"name": "a", "count": 5},
    {"name": "b", "count": 100},
    {"name": "b", "count": 32},
]

result = list_map_filter(my_list, filter_func=lambda x: x["count"] % 5 == 0, map_func=lambda x: x["name"])
# result is ["a", "b"]

result = list_map_filter(my_list, filter_func=lambda x: x > 50, map_func=lambda x: x["count"], filter_first=False)
# result is [100]

list_join

A helper function that joins a list with a given token. The Python default way for joining a list of items uses the join token (e.g. ",") and then calls .join on that string, passing the list of items as parameter. However, unfortunately this only accepts a list of strings and throws an error, if other types are passed (e.g. a list of numbers).

This helper casts all items to str before joining.

from pbu import list_join

my_list = ["a", 0, 4.5, False]

result = list_join(my_list, "-")
# result is "a-0-4.5-False"

result = "-".join(my_list)
# throws Error because my_list contains items of type other than `str`

not_none

A helper function to filter out None values from a list.

from pbu import not_none

my_list = ["a", None, "b", None, "c"]
result = not_none(my_list)
# result is ["a", "b", "c"]

Datetime Functions

PBU provides some utilities to help deal with timezones and datetime objects. All timezone specifications can be made either as a string (i.e. the name of the timezone, like "Australia/Melbourne") or as pytz.timezone object.

combine_date_time(date, time, tz)

Combines the provided date and time values.

from datetime import date, time
from pbu import combine_date_time

result = combine_date_time(date(year=2021, month=12, day=25), time(hour=15, minute=12, second=6), "Australia/Perth")

to_timezone(local_datetime, target_tz)

Translates a datetime to the provided target timezone.

from datetime import datetime
from pytz import utc
from pbu import to_timezone

utc_dt = datetime(year=2021, month=12, day=25, hour=3, minute=0, tzinfo=utc)  # 3:00am @ 2021-12-25
perth_dt = to_timezone(utc_dt, "Australia/Perth")
# > Result: 11:00am @ 2021-12-25 (+0800)

to_utc(local_datetime)

Shorthand for to_timezone(dt, pytz.utc)

set_timezone(datetime, target_timezone)

Simply replaces the timezone information without changing any of the time values of the datetime.

from datetime import datetime
from pytz import utc, timezone
from pbu import set_timezone

utc_dt = datetime(year=2021, month=12, day=25, hour=3, minute=0, tzinfo=utc)  # 3:00am @ 2021-12-25
perth_dt = set_timezone(utc_dt, timezone("Australia/Perth"))
# > Result: 3:00am @ 2021-12-25 (+0800)

weighted_mean

Provides the mean (average) of a list of values, where the values are weighted by the provided weights (in the same order as the value are provided). For missing weights, the default weight is 1

from pbu import weighted_mean

weights = [5, 3, 1]
values = [10, 5, 5, 4, 3]

# ((10 * 5) + (3 * 5) + (1 * 5) + 4 + 3) / (5 + 3 + 1) = 7.0
wm = weighted_mean(values, weights)  # 7.0

normalise

Normalises a numeric value between a lower and an upper boundary. The result is a value between 0.0 and 1.0. If the provided value exceeds any of the boundaries, the boundary value will automatically be chosen (defaults to 1.0 or 0.0).

It is possible to provide a smaller upper bound than lower bound, which will invert the function and provide the negated value. As an example, if we normalise 4 between 0 and 10, we get 0.4. If we invert the boundaries to normalise 4 between 10 and 0, we get 0.6 (1.0 - 0.4).

Any invalid input (None) will result in 0.0 being returned.

from pbu import normalise

# the "standard" case
norm1 = normalise(value=4, min_val=0, max_val=10)  # 0.4
# inverted normalisation
norm2 = normalise(value=4, min_val=10, max_val=0)  # 0.6
# exceeding the boundaries
norm3 = normalise(value=11, min_val=5, max_val=10)  # 1.0
# float works as well as integer
norm4 = normalise(value=-5.0, min_val=2.3, max_val=199.0)  # 0.0
# inverted exceeding boundaries
norm5 = normalise(value=-5, min_val=100, max_val=0.5)  # 1.0
# invalid inputs will return 0.0
norm6 = normalise(value=None, min_val=0, max_val=10)  # 0.0
norm7 = normalise(value=5, min_val=0, max_val=None)  # 0.0

Since version 1.0.1 a new parameter can be passed to the function that normalises the value, but can exceed the boundaries provided by min_val and max_val.

from pbu import normalise

norm1 = normalise(value=12, min_val=0, max_val=10, limit=False)  # 1.2

discretise

Discretises a numeric value into a number of buckets determined by the provided precision and boolean flag indicating whether to use the lower, upper or middle value of the bucket as the value for the bucket.

from pbu import discretise

disc1 = discretise(value=4.5, precision=1.0, floor=True)  # 4.0
disc2 = discretise(value=4.5, precision=0.4, ceil=True)  # 4.8
disc3 = discretise(value=4.5, precision=0.4)  # 4.6 (assumes mid-point if neither floor nor ceil is set)