Meta Learning for Code Summarization

A meta learning based approach for code summarization, is an attempt to combine the strengths of individual code summarization models through meta learning. See our paper for details:

@article{rauf2022meta,
  title={Meta Learning for Code Summarization},
  author={Rauf, Moiz and Pad{\'o}, Sebastian and Pradel, Michael},
  journal={arXiv preprint arXiv:2201.08310},
  year={2022}
}

This repository provides implementation for two neural network based models and one feature based model. This repository contains the following folder:

1. models: consisting of implementation for neural meta models and feature based model.
2. Scripts folder: containing utility scripts for computing output from meta models.

Training/Testing Models

We provide implementation for neural-meta models (Transformer and BiLSTM based) in addtion to feature based model. To perform training and evaluation, first go the scripts directory associated with the target model.

$ cd  models/MODEL_NAME

Where, choices for MODEL_NAME are ["feature_model", "neural_models"].

To train/evaluate a model, run:

$ bash run.sh 

Generated log files

While training and evaluating the models, a list of files are generated inside a results directory. The files are as follows.

• MODEL_NAME_epoch_id.pt
  • Model files containing the parameters of the model per epoch.
• Config.txt
  • Configuration file with hyperparameter details.
• predictions.json
  • The predictions file from the code. In addition to the files, a further res directory is created in order to save output of meta model and resultant bleu score. The following files are created in that folder.
• meta_summ.txt
  • File containing candidate summaries
• corpus_bleu.txt
  • File containing corpus BLEU.

Requirements

The code requires Linux and Python 3.6 or higher. It also requires installing PyTorch version 1.3 or higher. We additionally require NLTK and Scikit-learn. CUDA is strongly recommended for speed, but not necessary.