
This python utility package helps to create lazy modules.
A lazy module defers loading (some of) its attributes until these attributes are first accessed.
The module's lazy attributes in turn are attributes of other modules.
These other modules will be imported/loaded only when (and if) associated attributes are used.
A lazy import strategy can drastically reduce runtime and memory consumption.
Additionally, this package provides a utility for optional imports with which one can import a module globally while triggering associated import errors only at use-sites (when and if a dependency is actually required, for example in the context of a specific functionality).
lazy-imports
is available on the Python Package Index (PyPI).
[!WARNING]
Python's import system is highly complex and side effects are ubiquitous.
Although employing lazy imports (in a sanely structured project) is quite safe, you should keep in mind that there are necessarily subtle differences between lazy and ordinary (eager) imports/modules.
[!TIP]
Using a dedicated package such as this one means that you don't have to go through all the details yourself.
Still, we recommend to become acquainted with the basic functionality of lazy modules (such as understanding the roles of __getattr__
, __dir__
, and __all__
).
If you'd like to talk about it, feel free to open the discussion on Github.
Example 1
Type checkers cannot reason about dynamic attributes.
Therefore, we need a separate code path (using typing.TYPE_CHECKING
) on which regular imports are used.
By moving the imports to a separate file and using the module_source
utility, code duplication can be avoided.
from absolute.package import Class, function
from .submodule import Interface as MyInterface
[!IMPORTANT]
Lazy attributes have to be proper attributes (defined in __init__.py
) of the module they are imported from and cannot be submodules.
In this example, if function
is a submodule of absolute.package
, the lazy module won't (try to) load absolute.package.function
as a module by itself; in this case, an access of the lazy module's attribute function
might fail.
"""This is my great package."""
from typing import TYPE_CHECKING
from lazy_imports import LazyModule, as_package, load, module_source
__version__ = "0.1.0"
if TYPE_CHECKING:
from ._exports import *
else:
load(
LazyModule(
*as_package(__file__),
("__version__", __version__),
module_source("._exports", __name__),
name=__name__,
doc=__doc__,
)
)
Example 2
import ast as _ast
from typing import TYPE_CHECKING
from lazy_imports import LazyModule as _LazyModule
_mod = LazyModule(
*as_package(__file__),
"from . import echo",
ast.ImportFrom(names=[ast.alias(name="formats")], level=2),
ast.ImportFrom(module="filters", names=[ast.alias(name="equalizer", asname="eq")], level=2),
name=__name__,
)
if TYPE_CHECKING:
from . import echo
from .. import formats
from ..filters import equalizer as eq
else:
__getattr__, __dir__, __all__ = _mod.__getattr__, _mod.__dir__, _mod.__all__
try_import
is a context manager that can wrap imports of optional packages to defer exceptions.
This way you don't have to import the packages every time you call a function, but you can still import the package at the top of your module.
The context manager defers the exceptions until you actually need to use the package.
You can see an example below.
from lazy_imports import try_import
with try_import() as optional_package_import:
import optional_package
def optional_function():
optional_package_import.check()
optional_package.some_external_function()
[!TIP]
Instead of LazyImporter
we recommend using its successor LazyModule
, which
- allows attributes to be imported from any module (and not just submodules),
- offers to specify imports as plain python code (which can then be sourced from a dedicated file),
- supports
__doc__
, and
- applies additional sanity checks (such as preventing cyclic imports).
Usage example taken from hpoflow/__init__.py
:
import sys
from typing import TYPE_CHECKING
from lazy_imports import LazyImporter
from hpoflow.version import __version__
_import_structure = {
"mlflow": [
"normalize_mlflow_entry_name",
"normalize_mlflow_entry_names_in_dict",
"check_repo_is_dirty",
],
"optuna": ["SignificanceRepeatedTrainingPruner"],
"optuna_mlflow": ["OptunaMLflow"],
"optuna_transformers": ["OptunaMLflowCallback"],
"utils": ["func_no_exception_caller"],
}
if TYPE_CHECKING:
from hpoflow.mlflow import (
check_repo_is_dirty,
normalize_mlflow_entry_name,
normalize_mlflow_entry_names_in_dict,
)
from hpoflow.optuna import SignificanceRepeatedTrainingPruner
from hpoflow.optuna_mlflow import OptunaMLflow
from hpoflow.optuna_transformers import OptunaMLflowCallback
from hpoflow.utils import func_no_exception_caller
else:
sys.modules[__name__] = LazyImporter(
__name__,
globals()["__file__"],
_import_structure,
extra_objects={"__version__": __version__},
)
History
This project has previously been maintained by the One Conversation team of Deutsche Telekom AG.
It is based on _LazyModule
from HuggingFace and try_import()
from the Optuna framework.
Many thanks to HuggingFace for your consent
and to Optuna for your consent to publish it as a standalone package 🤗 ♥.
In December 2024 responsibility was transferred to Pascal Bachor.
Licensing
Copyright (c) 2024-2025 Pascal Bachor
Copyright (c) 2021 Philip May, Deutsche Telekom AG
Copyright (c) 2020, 2021 The HuggingFace Team
Copyright (c) 2018 Preferred Networks, Inc.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0.
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and limitations under the License.