Problems encountered in reproducing papers #2
Description
Dear Sir, I am a student who is studying cross project software defect prediction recently. Recently, I was carefully reading your paper Balancing Privacy and Utility in Cross Company Defect Prediction. I would like to ask about the difficulties encountered in the reproduction of a paper.
I am amazed at your important contribution in this direction, which has a great impact on my thinking. I am now trying to reproduce the effect in your paper, and I will use it according to your guidance website（ https://lace.readthedocs.io/en/latest/readme.html ）, use the same ten data sets in the paper, then use the naive Bayes based on Gaussian distribution and the naive Bayes based on polynomial distribution in sklearn, and use the default parameters, but the final results are poor. The g-measure of the classification performance evaluation of the data processed by LACE is far inferior to the original data. At present, every step of the operation has been checked and no problems have been found, so there is no idea for the time being.
After checking, LACE version is 2.1.2, and Python 2.7 is used when LACE is used. Python 3.8 is used when using sklearn for classification.
I would appreciate it if you could give me some valuable suggestions.