Negative p-values in VT results

[abought]

Reported by @gjmzajac :

The variable-threshold test in Raremetal is outputting a few hundred negative p-values. Is this a known bug?
For an example, see line 912 in the p-value column here is -302.613.

@welchr notes:

My guess is you're hitting what appeared to be a rarely reached piece of code in VT when the multivariate normal distribution function didn't work, and they had implemented some kind of alternative method. That code can return negative p-values in some cases.

raremetal/raremetal/src/Meta.cpp

Line 3775 in 2c82cfc

pvalue = CalculateMVTPvalue(score, cov_T, t_max);

First step would probably be to debug RAREMETAL and confirm where it is going wrong. I can do that if you would like, or I can help get you setup to do it, either way.

Greg has indicated he will produce a test case for this bug once other projects are done, and this ticket exists to track the known issue for other users in the interim.