The problem is that supplying cum_weights allows the code to run in O(log n) by using bisection. This is significantly faster on large populations. Adding a test that the cumulative weights are nondecreasing would add an O(n) step to the code.
So while I understand the OP's problem, I don't think it's soluble without making the cum_weights argument useless in practice. Better documentation might be worthwhile (although I don't personally find the current docs confusing, so suggestions for improvements would be helpful). Paul On 14 May 2018 at 12:36, Steven D'Aprano <steve+comp.lang.pyt...@pearwood.info> wrote: > Hi Paul, and welcome! > > On Sun, 13 May 2018 17:48:47 -0700, Paul wrote: > >> Hi, >> I just learned how to use random.choices(). > [...] >> Consequently, I specified 'cum_weights' with a sequence which wasn't in >> ascending order. I got back k results but I determined that they >> weren't correct (eg, certain population values were never returned). >> >> Since the non-ascending sequence, which I had supplied, could not >> possibly be valid input, why isn't this checked (and an error returned)? >> Returning incorrect results (which could be hard to spot as being >> incorrect) is much more dangerous. Also, checking that the list is in >> ascending order need only be done once, and seems like it would be >> inexpensive. > > Sounds like a reasonable feature request to me. > > > https://bugs.python.org/issue33494 > > > > -- > Steve > > -- > https://mail.python.org/mailman/listinfo/python-list -- https://mail.python.org/mailman/listinfo/python-list
