[BUG] bugfixes to QPD distributions - QPD_U, QPD_S

[fkiraly]

Re-enables tests for QPD_U, QPD_S skipped in #193 and attempts to fix failures in QPD distributions.

it appears that #190 failures are resolved - fixes #190.

Further QPD_B seems to fail at mean and var calculations, this is a separate bug and raised as #257. As a mechanical consequence, QPD_B is left skipped in the tests.

[fkiraly]

@setoguchi-naoki, @FelixWick, I have been trying to fix the open bug after re-enabling the tests, it seems like there are some issues with the internal indexing.

I'm not sure what format of alpha, qc_low etc the interna code assumes, could you help me? I was assuming, flat numpy array (1D), but that does not seem to work.

[FelixWick]

alpha: float,
qv_low: Union[float, np.ndarray],
qv_median: Union[float, np.ndarray],
qv_high: Union[float, np.ndarray],

[fkiraly]

I mean, internally. You do this: params[idx] = np.array([p]), so alpha becomes a 1D array, no?

[fkiraly]

Thanks, this seems to have helped.

Now, QPD_B is failing internally, it is substituting 0 or 1 into a logit, which produces nans.
Any ideas where or why, @FelixWick, @setoguchi-naoki?

It should not substitute the bounds, because they will always be nan, so where is that coming from?

[fkiraly]

FYI, it seems the bugs we were targeting are now resolved.

I have opened a bug issue here, to track that separately: #257