DomainAdaptedCodeT5

This repository is the repolication package for "An Empirical Investigation on the Performance of Domain Adaptation for T5 Code Completion".

How to run our code?

Every code is located in src/.

We use Pipenv as a virtual environment.

Create a dataset from a repository

pipenv run python src/create_dataset.py --model <path_to_model> --output <path_to_output>

Run a fine-tuning

pipenv run python src/training.py --model <path_to_model> --data <path_to_dataset> --output <path_to_output>

Without -model option, it is learned from the original CodeT5.

Run and measure

pipenv run python src/test.py --model <path_to_model> --data <path_to_dataset> --output <path_to_output>
pipenv run python src/analyze.py <path_to_csv>

test.py outputs the results of the model's predictions on the test data set as a csv.

analyze.py calculates various evaluation metrics from csv and adds them.