HD-GNN

For each commit, given its adjacency matrixs of entity reference graph and code change graph (Preprocessing code in https://github.com/fanmengdan/Preprocess_graph), this code aims at using HD-GNN to train a model which can classify the relations among code changes.

Due to the large size of the data file, we only present the preprocessed dataset of glide here(glide.zip needs to be decompressed). Please preprocess other data according to https://github.com/fanmengdan/Preprocess_graph

Requirements

• macOS/Windows/Linux
• anaconda (python 3.7)

Usage

1. Load adjacency matrixs of entity reference graph and code change graph of each commit.

• Enter adjacency matrixs data path in utils2.py：

repo = 'glide'
x = np.load(r'./Adjset/'+repo+'/Cutting_Adjs/CAdjs_' + str(step) + '.npy', allow_pickle=True)  # x是entity adj (100,200,200)  
y = np.load(r'./Adjset/'+repo+'/Cutting_Adjs/CHunkAdjs_' + str(step) + '.npy', allow_pickle=True)  # y是hunk adj (100,74,74)  

• Enter some file paths that can assist in generating intermediate products that HD-GNN can learn from in utils2.py：

IndexPathList = open(r'./dataset/'+repo+'/IndexPathList/IndexPathList_' + str(step) + '.pkl', 'rb')  
HunkIDmaps_path = open(r'./dataset/'+repo+'/HunkIDdict/HunkIDmap_' + str(step) + '.pkl', 'rb')  
IndexPaths = joblib.load(IndexPathList)  
HunkIDmaps = joblib.load(HunkIDmaps_path)

2. Train and Test

• Set model parameters in main.py, especially:

parser.add_argument('--Ne', type=int, default=entity_node, help='The Number of entities')  # 200, 250, 250  
parser.add_argument('--Nc', type=int, default=hunk_node, help='The Number of code changes')  # 74, 114, 150  
parser.add_argument('--Ner', type=int, default=entity_edge,  
 help='The Number of entity Relations')  # 39800, 62250, 62250  
parser.add_argument('--Ncr', type=int, default=hunk_edge,  
 help='The Number of code change Relations')  # 5402, 12882, 22350  
parser.add_argument('--Step', type=int, default=step, help='the number of commits/groups')  # 2, 3, 5  
parser.add_argument('--Repo', type=str, default='glide', help='the name of repository')

• Set up training or testing in main.py:

parser.add_argument('--Type', dest='Type', default='train', help='train or test')

Then run the main.py.