biolab / bottlechest

Fast NumPy array functions written in Cython

Bottlechest

Introduction

Bottlechest is a fork of bottleneck (https://github.com/kwgoodman/bottleneck), specialized for use in Orange (https://github.com/biolab/orange3).

Moving window functions, several other functions and all optimization of 3d arrays are removed to reduce the size of the library. New functions are added as needed.

NumPy/SciPy	median, nanmedian, rankdata, ss, nansum, nanmin, nanmax, nanmean, nanstd, nanargmin, nanargmax
Functions	nanrankdata, nanvar, replace, nn, anynan, allnan, nanequal

For other documentation, including a simple example and comprehensive set of benchmarks, refer to the original project.

License

Bottlechest is distributed under a Simplified BSD license. Parts of Bottleneck, NumPy, Scipy, numpydoc and bottleneck, all of which have BSD licenses, are included in Bottlechest. See the LICENSE file, which is distributed with Bottlechest, for details.

Install

Requirements:

Bottlechest	Python 2.6, 2.7, 3.2; NumPy 1.8
Unit tests	nose
Compile	gcc or MinGW
Optional	SciPy 0.8, 0.9, 0.10 (portions of benchmark)

bottlechest's Issues

bincount return type

Documentation for bincount says it returns a ndarray of ints. In fact it returns a ndarray of floats. As the values are sometimes used as indices this can be a problem.

I suppose floats are the correct choice, because the weighted version needs them and we wouldn't want different return types? In this case only the documentation needs to be changed, and casting to integers performed where needed (after using the function).

Pre rename check

Anything we want to simplify before we rename the project to bottlechest? We forked from bottleneck when it was version 0.6, now it is 0.8. Anything we want to include from their later releases?