daniel servén

@moptis have you had a chance to try this out again?

@moptis wow Im so sorry i havent answered yet! let me take another look!

@moptis grrr... hmm the p-values certainly look wrong, but are they still scaling wrt the data scale?

@alexgain i believe it :/ I frequently see these values as well. I'm not sure what to do until we fix this bug. - Should we remove the p-values from...

@shyamcody nice find. I took a look at the cross-validated answer, and it looks like we [accounted for the rank of the covariance matrix in the F-statistic](https://github.com/dswah/pyGAM/blob/b57b4cf8783a90976031e1857e748ca3e6ec650b/pygam/pygam.py#L1261). (also, I believe...

@shyamcody here is the [seemingly innocent line](https://github.com/dswah/pyGAM/blob/b57b4cf8783a90976031e1857e748ca3e6ec650b/pygam/pygam.py#L1040) where the covariance matrix is computed. The `B` matrix is computed in the PIRLS loop on https://github.com/dswah/pyGAM/blob/b57b4cf8783a90976031e1857e748ca3e6ec650b/pygam/pygam.py#L756 which corresponds to equation (4.32) on...

@xi2pi yes. to find prediction intervals for other distributions you should try the empirical sampling approach. ```python gam.fit(X, y) # your fitted model samples = gam.sample(X, y, quantity='y', n_draws=500, sample_at_X=X)...

please let me know how this works for you!

@conduit242 unfortunately, no. we need to have a fitted GAM first. we are drawing the samples from `p(y|X)` so we need to have fitted GAM coefficients that tell us how...

@xi2pi ahh i think the problem is mostly due to the fact that your X data are discrete, and thus you are generating multiple samples for many of your X...