Python package that can be used to enforce optional dependencies are installed when a module is imported. Necessary stands for "Now Ensures Correct Existence of Select Software, or Add Resource Yourself!"

Necessary is available on PyPI, and can be installed with the following command:

pip install necessary

Simply use necessary.necessary to get a context manager import a module.

from necessary import necessary

with necessary('torch'):
    # this will raise a nicely written error if
    # torch is not installed
    import torch

If you want to just soft fail when a necessary module is not available (that is, have necessary return False), use soft:

try:
    # assuming torch is not installed
    necessary('torch')
except ImportError:
    out = necessary('torch', soft=True)
    assert out is False

If you want to request a minimum version, use Python's requirements syntax:

with necessary('torch>=1.12.0'):
    # this will raise an error if
    # torch is not installed or if
    # the installed version is less than 0.12.0
    import torch

You can also check multiple packages in the same necessary call, or combine multiple requirements:

with necessary(['torch>=1.12.0', 'numpy>=1.20,<1.25']):
    # this will raise an error if torch >= 1.12.0 or numpy are not installed
    import torch
    import numpy

Finally, we can customize the message that is raised if a necessary module is not installed; use {module_name} and {module_version} to insert the module name and version respectively.

with necessary('torch', message='I am missing {module_name}/{module_version}'):
    import torch

You can also use necessary as a function or class decorator:

from necessary import Necessary

# decorating a function
@Necessary('torch')
def my_function():
    import torch

# decorating a class
@Necessary('torch')
class MyClass:
    def __init__(self):
        import torch

All of the same functionality is available in the decorator form as in the context manager form.