Typilus: Suggest Python Type Annotations

A GitHub action that suggests type annotations for Python using machine learning.

This action makes suggestions within each pull request as suggested edits. You can then directly apply these suggestions to your code or ignore them.

What are Python type annotations? Introduced in Python 3.5, type hints (more traditionally called type annotations) allow users to annotate their code with the expected types. These annotations are optionally checked by external tools, such as mypy and pyright, to prevent type errors; they also facilitate code comprehension and navigation. The typing module provides the core types.

Why use machine learning? Given the dynamic nature of Python, type inference is challenging, especially over partial contexts. To tackle this challenge, we use a graph neural network model that predicts types by probabilistically reasoning over a program’s structure, names, and patterns. This allows us to make suggestions with only a partial context, at the cost of suggesting some false positives.

How to use the Action in your Repository

To use the GitHub action, create a workflow file. For example,

name: Annotation Suggestions

# Controls when the action will run. Triggers the workflow on push or pull request
# events but only for the main branch
on:
  pull_request:
    paths:
      - '**.py'
    branches: [master, main]

jobs:
  suggest:
    # The type of runner that the job will run on
    runs-on: ubuntu-latest

    steps:
    # Checks-out your repository under $GITHUB_WORKSPACE, so that typilus can access it.
    - uses: actions/checkout@v3
    - uses: Karim-53/typilus-action@master
      env:
        GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
        SUGGESTION_CONFIDENCE_THRESHOLD: 0.8   # Configure this to limit the confidence of suggestions on un-annotated locations. A float in [0, 1]. Default 0.8
        DISAGREEMENT_CONFIDENCE_THRESHOLD: 0.95  # Configure this to limit the confidence of suggestions on annotated locations.  A float in [0, 1]. Default 0.95

The action uses the GITHUB_TOKEN to retrieve the diff of the pull request and to post comments on the analyzed pull request.

Technical Details & Internals

This GitHub action is a reimplementation of the Graph2Class model of Allamanis et al. PLDI 2020 using the ptgnn library. Internally, it uses a Graph Neural Network to predict likely type annotations for Python code.

This action uses a pre-trained neural network that has been trained on a corpus of open-source repositories that use Python's type annotations. At this point we do not support online adaptation of the model to each project.

Training your own model

You may wish to train your own model and use it in this action. To do so, please follow the steps in ptgnn. Then provide a path to the model in your GitHub action configuration, through the MODEL_PATH environment variable.

Contributing

We welcome external contributions and ideas. Please look at the issues in the repository for ideas and improvements.