Connecting Software Metrics across Versions to Predict Defects

SANER 2018, by Yibin Liu, Yanhui Li, Jianbo Guo, Yuming Zhou and Baowen Xu.

In this repository, we released the Matlab code for the RNN model for defect prediction in software engineering.

Format of HVSM:

For an HVSM connecting versions from v-T+1 to v, which has T versions counted, and there are n files in version v, the number of metrics is m. The input file has n rows, 1+(m+1)*T columns. Each row should be in the form of: version_length t, metrics in v-t+1, label in v-t+1, metrics in v-t+2, label in v-t+2, ..., metrics in v, label in v version_length t represent the number of versions that the file exists (t<=T). For those files with t < T, the columns should be filled with NaN in the tail of the row.

See rnnmain_test.m and our paper for more detailed information.

Usage

Just run rnnmain_test.m in Matlab environment.