Closed UnixJunkie closed 4 years ago
For example, iterate 50 times with molecules list randomization each time, train on 80%, test on the 20% left out test set. Then compare the distributions of their MCCs.
Yes, you can do that. Need to use a non-parametric method like Wilcoxon. In this case, I was focused on AUC. I have a blog post coming shortly that explains.
Ok, keep up with the super interesting chemoinformatics blog!
If you really compare their classification (not ranking) power.