Bottleneck

----------------

*Release date: 2012-06-04*

Thanks to Dougal Sutherland, Bottleneck now runs on Python 3.2.

**New functions**

- replace(arr, old, new), e.g, replace(arr, np.nan, 0)
- nn(arr, arr0, axis) nearest neighbor and its index of 1d arr0 in 2d arr
- anynan(arr, axis) faster alternative to np.isnan(arr).any(axis)
- allnan(arr, axis) faster alternative to np.isnan(arr).all(axis)

**Enhancements**

- Python 3.2 support (may work on earlier versions of Python 3)
- C files are now generated with Cython 0.16 instead of 0.14.1
- Upgrade numpydoc from 0.3.1 to 0.4 to support Sphinx 1.0.1

----------------

*Release date: 2011-06-13*

The fifth release of bottleneck adds four new functions, comes in a single
source distribution instead of separate 32 and 64 bit versions, and contains
bug fixes.

J. David Lee wrote the C-code implementation of the double heap moving
window median.

**New functions**

- move_median(), moving window median
- partsort(), partial sort
- argpartsort()
- ss(), sum of squares, faster version of scipy.stats.ss

**Changes**

- Single source distribution instead of separate 32 and 64 bit versions
- nanmax and nanmin now follow Numpy 1.6 (not 1.5.1) when input is all NaN

**Bug fixes**

- Support python 2.5 by importing `with` statement (:issue:`14`)
- nanmedian wrong for particular ordering of NaN and non-NaN elements (:issue:`22`)
- argpartsort, nanargmin, nanargmax returned wrong dtype on 64-bit Windows (:issue:`26`)
- rankdata and nanrankdata crashed on 64-bit Windows (:issue:`29`)

----------------

*Release date: 2011-03-17*

This is a bug fix release.

**Bug fixes**

- median and nanmedian modified (partial sort) input array (:issue:`11`)
- nanmedian wrong when odd number of elements with all but last a NaN (:issue:`12`)

**Enhancement**

- Lazy import of SciPy (rarely used) speeds Bottleneck import 3x

----------------

*Release date: 2011-03-08*

This is a bug fix release.

Same bug fixed in Bottleneck 0.4.1 for nanstd() was fixed for nanvar() in
this release. Thanks again to Christoph Gohlke for finding the bug.

----------------

*Release date: 2011-03-08*

This is a bug fix release.

The low-level functions nanstd_3d_int32_axis1 and nanstd_3d_int64_axis1,
called by bottleneck.nanstd(), wrote beyond the memory owned by the output
array if arr.shape[1] == 0 and arr.shape[0] > arr.shape[2], where arr is
the input array.

Thanks to Christoph Gohlke for finding an example to demonstrate the bug.

----------------

*Release date: 2011-03-08*

The fourth release of Bottleneck contains new functions and bug fixes.
Separate source code distributions are now made for 32 bit and 64 bit
operating systems.

**New functions**

- rankdata()
- nanrankdata()

**Enhancements**

- Optionally specify the shapes of the arrays used in benchmark
- Can specify which input arrays to fill with one-third NaNs in benchmark