BUG: Fixes gh-7491, pdf at x=0 of fisk/burr/burr12/f distributions. #10381
Conversation
Enable computation of pdf, logpdf at left-hand end points of fisk, burr, burr12 and f distrbutions. Fixes scipygh-7491 and half of scipygh-9544. Incorporate scipygh-9978 from rlucas7 to address scipygh-9544. Updated docstring of burr distribution. Added tutorial for stats.burr12. Removed an ignored implementation of fisk._stats(). In stats.burr.{pdf,logpdf} replaced two separate long code blocks with a simpler _lazywhere. Explicitly handle c<=n in stats.burr._munp(). Added checking of pdf-logpdf consistency to stats.test_continuous_basic.py. Added a filter for stats.recipinvgauss in check_pdf_logpdf(). Added any finite distribution endpoints to list of x-values checked in {check_pdf_logpdf, check_sf_logsf, check_cdf_logcdf}. Added another test case to test_burr_stats_9544().
| f2 = lambda x_, c_, d_: (c_ * d_ * (x_ ** (-c_ - 1.0)) / | ||
| ((1 + x_ ** (-c_)) ** (d_ + 1.0)))) | ||
| if output.ndim == 0: | ||
| return output[()] |
There was a problem hiding this comment.
This can be removed here, since this squeezing is done at the rv_generic.pdf anyway. Can be done as a follow-up though.
There was a problem hiding this comment.
Not all calls to ._pdf originate from inside .pdf(). rv_continuous._cdf_single() calls integrate.quad(self._pdf, ...).
There was a problem hiding this comment.
Many other distribution do not have this clause. Is it a bug in them all? Or is it a bug in _cdf_single?
There was a problem hiding this comment.
integrate.quad passes in scalars to self._pdf and expects a scalar back.
burr._pdf(self, x, c, d) has a conditional. It does this by invoking _lazywhere(...), and _lazywhere always returns an array, even if all its inputs are scalars. Hence the squeeze.
The _pdf of most distributions makes calls to numpy or scipy ufuncs which return a scalar if given scalars, so this issue doesn't affect them.
| + sc.xlogy(-c_ - 1, x_) | ||
| - sc.xlog1py(d_+1, x_**(-c_)))) | ||
| if output.ndim == 0: | ||
| return output[()] |
|
Thanks @pvanmulbregt, @rlucas7, LGTM modulo nitpicks which can be addressed in a separate cleanup. Merged. |
Replaces gh-9983 (rebase had trouble merging with gh-10154 --- this PR applies the same modifications manually to current master.)
Enable computation of
pdf,logpdfat left-hand end points offisk,burr,burr12andfdistrbutions. Fixes gh-7491 and half of gh-9544.Incorporate gh-9978 from @rlucas7 to address gh-9544.
Updated docstring of
stats.burrdistribution.Added tutorial for
stats.burr12.Removed an ignored implementation of
fisk._stats().In
stats.burr.{pdf,logpdf}replaced two separate long code blocks with a simpler_lazywhere.Explicitly handle
c<=ninstats.burr._munp().Added consistency checking of
pdfandlogpdftostats.test_continuous_basic.py.Added a filter for
stats.recipinvgaussincheck_pdf_logpdf().Added any finite distribution endpoints to list of x-values checked in
{check_pdf_logpdf, check_sf_logsf, check_cdf_logcdf}.Added another test case to
test_burr_stats_9544().