• Python 2.7
• Tensorflow 1.12.0

Python and JavaScript datasets: https://plml.ethz.ch

Java dataset: https://drive.google.com/open?id=1xxnYAu8L5i6TpNpMNDWxSNsOs3XYxS6T

Each program is represented in its AST format, and the AST is serialized in in-order depth-first traversal to produce the AST node sequence.

Data process code is in the "preprocess_code" diretory.

python train_gpu_mt.py --alpha ${weight for type prediction loss} --mem_len ${memory length} --model_dir ${path_to_save}