Gpv2DP

Software Defect Prediction via Code Grayscale Pixel Visualization with Fusion Attention(Gpv2DP).

We proposed a novel Gpv2DP approach, which simultaneously considers code visualization comprehension and fusion attention mechanism to extract the code feature for software defect prediction.

Prepare for dataset

• Put the source code of the experiment project in folder ./data/archives. Available via github or other websites.
• Generate the code images corresponding to the java files. Please run make_txt.py, code_vis.py, and makeInstance_txt.py sequentially.
• The path structure of the prepared data images is as follows:

-data
  -archives
  -csvs
  -img
    - Project 1
      -file 1.png
      -file 2.png
      ...
    - Project 2
      -file 1.png
      -file 2.png
      ...
  -txt

Build running environment

• Install required packages.

pip install -r requirements.txt

• Install accimage (support Windows, Linux and macOS). Official link: https://github.com/pytorch/accimage

$ conda install -c conda-forge accimage

Train and test

• To train and test, simply run train.py.

python train.py

Config batch training

• Modify and execute the run.sh.

sh run.sh

Check experimental result

• Please check the folder ./temp/result/.

Contacts

• Due to anonymity requested, we will publish contact details in the future.