ffl's Introduction

FFL

Dataset Downloads

Download dataset Codeflaws here: https://codeflaws.github.io
Download dataset Prutor here: https://bitbucket.org/iiscseal/nbl/src/master/

Extract the 2 dataset into data_codeflaws and data_nbl respectively

Environment:

The following dependencies are required to train or run the model:

pytorch (1.6.0+)
fasttext (0.9.2)
dgl (1.6.0)
networkx (2.5.1)

Optional (for visualization):

pygraphviz and graphviz

Java: version 1.11+

Gumtree-based annotation package: https://drive.google.com/file/d/1hQ5wmSs2K6dPdhC38VUzn-dtK3J19LDf/view?usp=sharing

Extract jars.zip into folder jars

Training:

# Codeflaws node-level
python3 -m codeflaws.train_nx_a_nc_cfl
# NBL node-level
python3 -m nbl.train_nx_a_nc

# Codeflaws statement-level
python3 -m codeflaws.train_nx_astdiff_nocontent_gumtree
# Prutor statement-level
python3 -m nbl.train_nx_astdiff_nocontent_gumtree

Evaluation

The training script already contain evaluation, by disabling train() procedure, the script will script directly to evaluation. Copy the pretrained model into train_dirs in the configured utils/utils.py for evaluate.

Pretrained model:

Codeflaws node-level objective: https://drive.google.com/file/d/1E0XHg5J3wBFaGh8DYL4DmCNf-0z9z58_/view?usp=sharing
Codeflaws statement-level objective: https://drive.google.com/file/d/1D6_YrUDCfdYqgsRwvpRA-le_MHDrFMhV/view?usp=sharing
Prutor node-level objective: https://drive.google.com/file/d/1kfL8TpxYA491PUyKvd4nzpbFwcfqGGyY/view?usp=sharing
NBL statement-level objective: https://drive.google.com/file/d/1Da00veDZb-445eruuE3boOoxzl1twak8/view?usp=sharing

Others

Please note that while this is not required in our original settings, codebert pretrain file can be placed in preprocess folder for each AST content to be used instead of just nodetype.

Recommend Projects