https://github.com/tillahoffmann/localscope

Restrict the scope of functions for reproducible code execution and peace of mind.
https://github.com/tillahoffmann/localscope

jupyter machine-learning python reproducibility research

Last synced: 10 days ago
JSON representation

Restrict the scope of functions for reproducible code execution and peace of mind.

• Host: GitHub
• URL: https://github.com/tillahoffmann/localscope
• Owner: tillahoffmann
• License: mit
• Created: 2020-10-14T21:49:33.000Z (about 4 years ago)
• Default Branch: main
• Last Pushed: 2024-07-15T19:48:53.000Z (4 months ago)
• Last Synced: 2024-10-12T18:30:55.009Z (26 days ago)
• Topics: jupyter, machine-learning, python, reproducibility, research
• Language: Python
• Homepage: https://localscope.readthedocs.io
• Size: 37.1 KB
• Stars: 4
• Watchers: 3
• Forks: 1
• Open Issues: 1
• Metadata Files:
  • Readme: README.rst
  • License: LICENSE

Awesome Lists containing this project

README

        
🔐 localscope

=============

.. image:: https://github.com/tillahoffmann/localscope/actions/workflows/build.yml/badge.svg

  :target: https://github.com/tillahoffmann/localscope/actions/workflows/build.yml

.. image:: https://img.shields.io/pypi/v/localscope.svg

   :target: https://pypi.python.org/pypi/localscope

.. image:: https://readthedocs.org/projects/localscope/badge/?version=latest

  :target: https://localscope.readthedocs.io/en/latest/?badge=latest

Interactive python sessions, such as `Jupyter notebooks `__, are outstanding tools for analysing data, generating visualisations, and training machine learning models. However, the interactive nature allows global variables to leak into the scope of functions accidentally, leading to unexpected behaviour. Localscope gives you peace of mind by restricting the variables a function has access to.

.. doctest::

   >>> a = 'hello world'

   >>> @localscope

   ... def print_a():

   ...     print(a)

   Traceback (most recent call last):

     ...

   localscope.LocalscopeException: `a` is not a permitted global (file "...",

      line 1, in print_a)

Motivation and detailed example

-------------------------------

Suppose you are evaluating the mean squared error between two lists of numbers, including a scale factor ``sigma``.

.. doctest::

   >>> sigma = 7

   >>> # [other notebook cells and bits of code]

   >>> xs = [1, 2, 3]

   >>> ys = [4, 5, 6]

   >>> mse = sum(((x - y) / sigma) ** 2 for x, y in zip(xs, ys))

   >>> mse

   0.55102...

Everything works nicely, and you package the code in a function for later use but forget about the scale factor introduced earlier in the notebook.

.. doctest::

   >>> def evaluate_mse(xs, ys):  # missing argument sigma

   ...     return sum(((x - y) / sigma) ** 2 for x, y in zip(xs, ys))

   >>> mse = evaluate_mse(xs, ys)

   >>> round(mse, 5)

   0.55102...

The variable ``sigma`` is obtained from the global scope, and the code executes without any issue. But the output is affected by changing the value of sigma.

.. doctest::

   >>> sigma = 13

   >>> evaluate_mse(xs, ys)

   0.15976...

This example may seem contrived. But unintended information leakage from the global scope to the local function scope often leads to unreproducible results, hours spent debugging, and many kernel restarts to identify the source of the problem. Localscope fixes this problem by restricting the allowed scope.

.. doctest::

   >>> from localscope import localscope

   >>> @localscope

   ... def evaluate_mse(xs, ys):  # missing argument sigma

   ...     return sum(((x - y) / sigma) ** 2 for x, y in zip(xs, ys))

   Traceback (most recent call last):

     ...

   localscope.LocalscopeException: `sigma` is not a permitted global (file "...",

      line 3, in )

Interface

---------

.. autofunction:: localscope.localscope