Supplementary code and data of the paper Evaluating network embedding techniques' performances in software bug prediction.

Supplementary code and data of the paper Evaluating network embedding techniques' performances in software bug prediction.

This repo can also be used as a complementary material for our previous paper:

Qu, Yu, Ting Liu, Jianlei Chi, Yangxu Jin, Di Cui, Ancheng He, and Qinghua Zheng. "node2defect: using network embedding to improve software defect prediction." In 2018 33rd IEEE/ACM International Conference on Automated Software Engineering (ASE), pp. 844-849. IEEE, 2018. https://dl.acm.org/doi/abs/10.1145/3238147.3240469

Steps to Run this dataset:

First: Generating Class Dependency Network

In each subdirectory, we have already included the corresponding Class Dependency Network (CDN) (classgraph.dot). If you want to generate your own CDN, you can use the Understand Perl Script file -- Class-Graph.pl, after installing the Understand tool, by using the commend:

uperl Class-Graph.pl %YourOwnProjectDirectory%

Second: Generating the input file for network embedding algorithms

After generating the CDN, we can use the Driver.py to generate the input file for network embedding algorithms.

python Driver.py

After executing Driver.py, we can get the "edgelist" file in each directory. Then, for instance, we can use the ProNE implementation to generate the embedding file (classgraph.emd):

python proNE.py -graph edgelist -emb1 classgraph.emd -emb2 classgraph-2.emd -dimension 32 -step 10 -theta 0.5 -mu 0.2

Third: Run the experiment

After generating the classgraph.emd file, we can run the experiment by executing Node2Defect-Final-CrossValidation.py:

python Node2Defect-Final-CrossValidation.py

The overall results are listed in the files like "All-Popt-%s-%s-%s.csv"

Requirements:

python==3.7
scipy==1.5.2
networkx==2.5
scikit-learn==0.23.2
numpy==1.19.1
pandas==1.1.2

We have used the following implementations of different network embedding algorithms, please refer to their repos for their dependencies:

OpenNE
ProNE
Walklets