It seems to work well (see the tests) for both geo pooled and non geo pooled (more generally key col pooled) cases, but I still don't know how to keep this from running at test time, where the number of data points available will be way different and may cause problems.
Attempt at fixing #106.
It seems to work well (see the tests) for both geo pooled and non geo pooled (more generally key col pooled) cases, but I still don't know how to keep this from running at test time, where the number of data points available will be way different and may cause problems.
@dajmcdon any suggestions?