Deep Learning-based type inference for Python

This repository contains instructions for using deep learning-based approaches that predicts types for Python.

Requirements

To install required dependencies for the approaches, run the following command:

pip install -r requirements.txt

Datasets

Approaches

TypeWriter

The TypeWriter is proposed by:

Pradel, M., Gousios, G., Liu, J., & Chandra, S. (2019). TypeWriter: Neural Type Prediction with Search-based Validation. arXiv preprint arXiv:1912.03768.

Script

The re-implementation of this approach consists of two parts:

• Extractor: To process functions and generate data vectors, use the following script:

  python TW_extractor.py --o $OUTPUT_FOLDER --d $REPOS --w $THREADS

  $OUTPUT_FOLDER: Specify a folder to store the output files. The folder will be created automatically.

$REPOS: The path to Python projects

$THREADS: Number of workers for extracting functions from the projects.

• Model: To learn the neural model and do prediction, run the following script:

  python TW_model.py --o $OUTPUT_FOLDER

  $OUTPUT_FOLDER: Note that this is the output folder that you specified in the extractor part.

To change the hyper-parameters of the neural model, you can change the values in the file data/tw_model_learning_params.json.

It should be noted that both Extractor and Model scripts save the required files for running the inference script in $OUTPUT_FOLDER/tw_model_files.

Notebook

To run all the steps of TypeWriter manually with explanation, check out the notebook main_TW.ipynb.

Inference

By employing the pre-trained neural model of TypeWriter, you can infer both argument and return types of the methods of a given Python source file using the following script:

python TW_inference.py --s $SRC_FILE --m $TW_MODEL_PATH --t $TOP_K

$SRC_FILE: A Python source file.

$TW_MODEL_PATH: The path to the TypeWriter's pre-trained model and auxiliary files.

$TOP_K: Top K predictions (the default value of K is 3)