Conversation
JTFouquier
left a comment
There was a problem hiding this comment.
This is an improvement, thanks!
The bigger picture that I didn't address well is for cases when one model works completely (like Original or Pairwise, which is common), but when the other models like First or Previous fail for the RF/MERF step or BorutaSHAP step. The report should still be created because one model worked and there should be no results in the other sections. When RF fails at <5% percent variation explained, I made empty PDFs to "fail elegantly"... this can potentially be reused for the BorutaSHAP step. Maybe I just didn't think of that case.
Cases could be as follows as far as "selecting features" for each model (Original, First, Previous or Pairwise):
RF/MERF pass
BorutaSHAP pass
This builds complete pdf reports; satisfies pdf file for Snakefile rule all
RF/MERF pass
BorutaSHAP fail
Currently causing issues especially when some models works but not others; report won't be created here.
RF/MERF fail
BorutaSHAP fail
Fails at RF/MERF so an empty pdf is created to satisfy rule all, so BorutaSHAP was irrelevant here.
Now that I wrote this out, I'll give it a go. :)
Hi @JTFouquier. I started this pull request to address Jaime's problem (see Issue #3 ). This only addresses one particular case though. I don't know if you want to cover other possible failure cases.