Evaluation Metric of WxBS Benchmark about roma

[Parskatt]

Yes, that should be right. The results should be exactly the ones reported by Mishkin, so if you have further questions about the metrics, please ask him. I remember when rerunning the experiment that the results can fluctuate by about 1%

[cailile]

Thanks for the prompt reply. I did observe some fluctuation for different runs and the best one I get is 79.97%, a little less than the one reported in your paper (80.1). May I know in such case you just report the best one or the average of multiple runs? What is the common practice?

[Parskatt]

Reporting the average I think is reasonable. Otherwise you can increase score by doing more runs :D

[Parskatt]

Its likely though that my original numbers are from a single run.