Add CI calculation over posterior samples for a percentage correct other than default #242
Description
@guillermoaguilar thanks for the careful explanation, I understand the argument now. We are not computing a real CI over the threshold, but this is what the warning is about.
The alternative way to compute the CI using the samples would marginalize over all the other parameters and indeed it could give an accurate (if non-deterministic) estimate of the CI over the threshold.
I don't have objections with going forward with the implementation in #241. Should we complete that PR and then discuss how to implement a sampling-based CI?
Originally posted by @pberkes in #191
We do this inside method threshold( ) in the results object. We need to run enough posterior samples and from them get percentile CIs for the percentage correct asked.
Also need to update the documentation page.