README

Content

• dependency_analyzer.py: load dependency analyzer and estimate dependency score;
• siamise_model.py: load, train and inference semantic similarity model;
• main.py: train semantic similarity model and fit a linear regression model to estimate edit propagation score of a file;
• dataset/: dataset for training and evaluation, available in Huggingface;
• models/: semantic embedding models, available in Huggingface.

Key results

Language	Accuracy	Precision	Recall	F1
JavaScript	88.77%	81.52%	71.21%	76.02%
Python	85.78%	70.84%	73.40%	72.09%
Java	90.65%	85.28%	75.67%	80.19%
Go	88.55%	80.10%	72.12%	75.90%
Typescript	88.46%	79.84%	72.25%	75.86%
Average	88.44%	79.52%	72.93%	76.01%

Usage

• Adjust hyperparameters in main.py and run python main.py to train semantic similarity model and fit a linear regression model to estimate edit propagation score of a file.