nashid / ai-program-augmentation

Project: 2020SU-COSC6368-Artificial Intelligence

AI-Program-Augmentation

Project: 2020SU-COSC6368-Artificial Intelligence
Task: Apply Data Augmentation in Program Domain.

---

Library:

• Python 3.7.4
• javaparser 3.14.5

---

Program Augmentation:

Given input and output path, create jar and execute jar.

• input = Input directory to the original programs.
• output = Output directory to the augmented programs.

$ cd augment/
$ source run.sh
$ java -jar JavaAugmentation/target/jar/JavaAugmentation.jar $input $output

Types:

• Function Augmentation
• Statement Augmentation
• Loop Augmentation
• Switch Augmentation
• Binary Augmentation

Example:

( An example of Function Augmentation (left) and Statement Augmentation (right). Check here for other types. )

Results

Dataset	Partition	Best Epoch	Accuracy (%)
Original	Validation	7	87.11
Augmented	Validation	8	90.70
Original	Test	7	90.81
Augmented	Test	8	93.35

References:

• code2vec: https://github.com/tech-srl/code2vec
• SA: https://github.com/bdqnghi/ggnn.tensorflow#code-classification
• javaparser: https://github.com/javaparser/javaparser