Repository for paper "Improving Transformer-based Program Repair Models through False Behavior Diagnosis"
Here we show example code and data used for the experiment.
[Folders]
A. APR Models
- Trained APR models
- Trained models can be found in following link:
https://www.dropbox.com/sh/q94kdtguedgl4f7/AACSMqydlIoQVqVSI91DTTZIa?dl=0
B. Attentions
- Sample normalized attention maps used for experiment
- Pickle files also contain other infos such as input_id
- Entire pickle files for CodeT5 Model with Wild-small^na Dataset can be found in following link:
https://www.dropbox.com/sh/q94kdtguedgl4f7/AACSMqydlIoQVqVSI91DTTZIa?dl=0
C. Dataset
- Buggy dataset (buggy code - fixed code pairs)
D. Extracted Attentions
- Temprorary folder to show how we extracted normalized attention map
- "1. Collect Normalized Attention.ipynb" shows the extraction process
[Files]
1. Collect Normalized Attention.ipynb
2. Diagnose and T2 Example.ipynb
3. transformers.zip
- We used the modified transformer==4.12.3 for our experiment
4. README.txt
5. requriements.txt