Network-clustered Multi-modal Bug Localization.

Building the NetML algorithm, the figure belows illustrate our proposed model:

First time setup

Please install the neccessary libraries before running our NetML model:

• python 2.7

• numpy 1.13.1

• scikit-learn 0.19.0

Dataset details

• The dataset is put in two folders ./data and ./data_defect4j. The folder ./data includes four projects, namely Ant, Aspectj, Lucene, and Rhino, whereas the folder ./data_defect4j includes three projects, namely Lang, Math, and Time.

Example running

• Please run test_NetML.py to get used to NetML model. We see that the loss value descrease which means that our loss function converges.

Input data

• All the bug reports for each project.

• All the methods for each project.

• Features for each pair of bug report-method, the features mention whether a relationship between the bug report and method.

• Label data

• Please take a look at folder ./data_example to see the input of our framework:

  • bug_list.txt: all the bug reports
  • method_list.txt: all the methods
  • features.txt: features for each pair of bug-method
  • groundtruth.txt: label data

Parameters:

We have five different parameters:

• nfolds: number of folds to do cross-validation
• iters: number of iterations for training NetML
• alpha and beta: control the strength of ridge and Network lasso regularization, respectively.
• kNN: number of nearest of neighbors

Running the model

Simply run this command to train the network:

$ python run_NetML.py ./data_example/bug_list.txt ./data_example/method_list.txt ./data_example/features.txt ./data_example/groundtruth.txt 10 30 0.1 0.01 10

Note that in this case, nfolds=10, iters=30, alpha=0.1, beta=0.01, and kNN=10

Example output: