This allows quantiles of 0 and 1 to be used for qstudent_t, qskew_normal, qfrechet, qgen_extreme_value and qasym_alplace.
Check if you like it the way I did. In addition, rather than throwing an error when p <0 or >1, the functions produce NaN like all other q* type distribution functions. It probably rarely matters, but I think the consistency is nice. The check itself is still necessary, because some distributions like qfrechet would produce non-meaningful but real values for p > 1.
I added a couple of new tests to check the indended results with p at or beyond the limits
Incorrect output of qgen_extreme_value
While working on the above, I noticed that qgen_extreme_value produces incorrect results when xi != 0. There was no test for qgen_extreme_value, so I initially added the following test (similar to what was there for qfrechet):
Closes #1637
This allows quantiles of 0 and 1 to be used for qstudent_t, qskew_normal, qfrechet, qgen_extreme_value and qasym_alplace.
Check if you like it the way I did. In addition, rather than throwing an error when p <0 or >1, the functions produce NaN like all other q* type distribution functions. It probably rarely matters, but I think the consistency is nice. The check itself is still necessary, because some distributions like qfrechet would produce non-meaningful but real values for p > 1.
I added a couple of new tests to check the indended results with p at or beyond the limits
Incorrect output of qgen_extreme_value
While working on the above, I noticed that qgen_extreme_value produces incorrect results when xi != 0. There was no test for qgen_extreme_value, so I initially added the following test (similar to what was there for qfrechet):
This line was responsible, so I fixed it and now the new test passes. Let me know if you prefer to have this fix in a separate PR.