-
Tensorflow-gpu == 1.0.0
-
Python 3.6
-
CUDA 8.0 (For GPU)
This is the implementation of the paper "Hierarchical Embedding for Code Search in Software Q&A Sites".
The dataset can be obtained from here: https://figshare.com/s/f1e97bfde7b7b5960d4a
-
CodeATTENTION.py : The implementation of CodeATTENTION.
-
CodeONLSTM.py : The implementation of CodeONLSTM.
-
HECS.py : The implementation of HECS.
-
attention_visualization.py : The implementation of drawing heatmap of attention.
-
data_helper.py : Data helpy function for this experiments.
-
main.py ：The Main function of this experiments.
-
model01_base_LSTM_CNN.py ：The implementation of CodeRCNN
-
model02_base_CNN.py ：The implementation of CodeCNN
-
model03_base_LSTM.py ：The implementation of CodeLSTM
-
model04_base_CODEnn.py ：The implementation of DeepCS
-
model_utils.py The metrics of experiments
We provide example codes to repeat the experiments.
$ python main.py --train=True
Selecting this parameter means training the model.The results (Recall@1 and Mean Reciprocal Rank (MRR)) of each training epoch are given during the training. If you want to get other results (FRank, Precision@k) you need to modify the evaluation function (eval_metric_rec1) of the main.py to "eval_metric".
$ python main.py --pred=True
Selecting this parameter means testing the existing model.The results (Recall@1 and Mean Reciprocal Rank (MRR)) of each training epoch are given during the training. If you want to get other results (FRank, Precision@k) you need to modify the evaluation function (eval_metric_rec1) of the main.py to "eval_metric".
$ python main.py --pmap=True
Selecting this parameter means drawing heatmap of attention based on existing models.