problem Level
rate 1>=x > 2/3 easy 650 rate 2/3>=x > 1/3 normal 158 rate 1/3>=x >= 0 hard 322
LevelSplit
According to Problem Level, Split Problem to different level.
Easy_count = 650 Normal_count = 158 Hard_count = 322
Train PId count = 906 150 Dev PId count = 111 10 Test PId count = 113 20
Token 2048 record count = 65355->64538 record count = 949->927 record count = 1767 ->1688
finished processing 64538 data. in ./repairDataset/RepairData-PythonLevel/train.json loading dataset from: ./repairDataset/RepairData-PythonLevel/dev.json finished processing 927 data. in ./repairDataset/RepairData-PythonLevel/dev.json loading dataset from: finished processing 1688 data in ./repairDataset/RepairData-PythonLevel/test.json
bug localization (fault localization)
If we only consider that the data's token is less than 1024, the number of training data points is 50,223.
datasets
CRFLPDataset (Code trace fault localization Problem)
used for trianing Bug Locator and test result
mixCRPLPDataset (Mixed Code trace fault localization Problem)
used for training Program Modifier with Hybrid Training for Selective Reference
Preference training data
"dataNew/FixPerferdataset" used for training Program Modifier with Adaptive Preference Learning