This repo provides the code for reproducing the experiments in ASTBERT: A Fine-tuned BERT model for Programming Languages based on ASTs. Unlike many other AST-based tools ASTBERT uses the subtrees of ASTs to extract syntactical information from programming languages.

For code classification, unzip and release the csv training data in folder 'data'.

For code clone detection untar and release the Json file in folder 'data'.

tar -xvf data_Java_code_pairs.tar

pip install keras pip install transformers pip install tensflow==2.2 pip install keras-bert

run python train.py

run python train_code_clone_detection.py