I’m going to strongly support David Mertz’s point here: It is a well known axiom of computing that returning an *incorrect* result is a very bad thing.
What the correct result of the median of a sequence of floats that contains some NaNs is up for debate. As David points out there are (at least) three reasonable answers: - NaN - the median that ignores the NaNs - an Exception But having a result that essentially treats NaN as having an arbitrary value depending on where it is the sequence is not correct by any definition. And while it’s nice that median supports duck typing, floats are probably the most (certainly the second most) common type used with it — some special case code is called for if needed. Python is not a high performance computational language. If you are crunching enough numbers that a NaN check is going to make a difference, you probably should be using numpy or some other higher performance stats lib. So I think this should be addressed, and it's well worth a performance hit to do so. Which does bring us to how to do it. I think there are two possible approaches: 1) do checks in the median function itself -- I'm going to suggest that that's the place to do it -- and indeed, probably a good idea to review the statistics module for other places where NaNs will cause problems. 2) in the sort function(s) -- it would b nice if sort did something "smarter" with NaNs, but if that smarter thing is something other than a Exception, median() should still check for NaNs; if they are al at the end of beginning, you will at least get consistent results, but still not really meaningful ones -- I think NaNs should either be treated as missing values, or not allowed at all. NOTE: a while back someone was suggesting that the sort function(s) check for simple C datatypes in the keys -- and if they are, e.g. floats or ints, use fast C code for the sorting -- he claimed it would make sorting much faster for these common cases. I don't know what came of that, but if it is implemented, then adding NaN handling there might make some sense. -CHB
_______________________________________________ Python-ideas mailing list -- python-ideas@python.org To unsubscribe send an email to python-ideas-le...@python.org https://mail.python.org/mailman3/lists/python-ideas.python.org/ Message archived at https://mail.python.org/archives/list/python-ideas@python.org/message/ENCEOW4QVETPCIEQ5VE2ERTJEXHSDFDG/ Code of Conduct: http://python.org/psf/codeofconduct/
