Adding a boltons.statsutils.mode() function about boltons

[mahmoud]

Sure! Sounds like a fine enhancement. Would you care to give it a go? Also, I'm over here wondering if a bucketized version might help make it more useful for float variables, or do you think that would make it more confusing?

[ajmazurie]

I would restrict a mode function to discretized values (integers or strings). What you suggest sounds more like a histogram function, such as numpy.histogram where the user has to say if two values (such as floats) belong to the same bucket or not, either by requesting a given number of buckets or requesting a given bucket width.

I would be more than happy to contribute a patch (never done it before, so bear with me!)

[tiwo]

Python >=3.4 comes with the statistics module, providing median, mean, mode, harmonic mean (mentioned in boltons/TODO.rst), variance/standard deviation.

Maybe this means that providing these functions in boltons isn't as urgent, or maybe it means that they can just be backported / copy-and-pasted from the standard library (I believe the Python license is compatible).

[mahmoud]

Hmm, the 3.4+ stdlib version is pretty simplistic. It doesn't really encapsulate the dataset-based approach (with caching, etc.). It also inexplicably introduces the complexity of sample-based statistics by applying Bessel's correction all over the place, when most simple applications are safe to presume population-based approaches. But, after stripping away those two design oversights, if you still find something worth cribbing, would be happy to include it! I still hear from many Python <3.4 boltons users all the way back to Python 2.6, and they're always happy to have more features. :)