Bottleneck is a collection of fast NumPy array functions written in Cython:
NumPy/SciPy |
|
| Functions | nanrankdata, nanvar, partsort, argpartsort |
Moving window |
|
Let's give it a try. Create a NumPy array:
>>> import numpy as np
>>> arr = np.array([1, 2, np.nan, 4, 5])Find the nanmean:
>>> import bottleneck as bn
>>> bn.nanmean(arr)
3.0Moving window nanmean:
>>> bn.move_nanmean(arr, window=2)
array([ nan, 1.5, 2. , 4. , 4.5])Bottleneck is fast:
>>> arr = np.random.rand(100, 100)
>>> timeit np.nansum(arr)
10000 loops, best of 3: 82.8 us per loop
>>> timeit bn.nansum(arr)
100000 loops, best of 3: 16.9 us per loopLet's not forget to add some NaNs:
>>> arr[arr > 0.5] = np.nan
>>> timeit np.nansum(arr)
10000 loops, best of 3: 126 us per loop
>>> timeit bn.nansum(arr)
100000 loops, best of 3: 63.1 us per loopBottleneck comes with a benchmark suite. To run the benchmark:
>>> bn.bench(mode='fast', dtype='float64', axis=0)
Bottleneck performance benchmark
Bottleneck 0.5.0
Numpy (np) 1.6.0
Scipy (sp) 0.9.0
Speed is NumPy or SciPy time divided by Bottleneck time
NaN means one-third NaNs; float64 and axis=0 are used
High-level functions used (mode='fast')
no NaN no NaN no NaN NaN NaN NaN
(10,10) (100,100) (1000,1000) (10,10) (100,100) (1000,1000)
median 5.12 2.12 2.23 5.54 4.48 2.95
nanmedian 115.31 28.86 4.23 140.78 70.29 6.49
nansum 10.14 4.88 1.69 10.21 5.75 1.68
nanmax 2.26 1.31 1.01 2.51 3.98 1.09
nanmean 25.02 10.20 2.38 26.41 25.22 4.41
nanstd 32.22 7.47 2.34 33.76 15.43 3.38
nanargmax 10.25 5.48 2.59 10.63 8.10 2.78
ss 5.51 2.55 1.29 5.48 2.62 1.23
rankdata 22.65 13.36 8.81 23.02 14.44 9.91
partsort 1.52 2.13 2.43 1.75 5.96 3.79
argpartsort 0.77 2.15 1.68 0.75 3.71 1.70
move_sum 10.52 8.33 14.25 10.24 8.90 13.84
move_nansum 25.53 20.06 29.20 26.02 25.46 29.38
move_mean 10.84 4.42 14.38 11.10 8.82 13.93
move_nanmean 32.50 11.64 29.74 34.26 14.34 30.49
move_std 17.77 3.43 22.94 23.19 21.07 29.77
move_nanstd 35.07 6.03 34.87 40.64 6.92 36.10
move_max 4.32 3.73 9.32 4.77 5.86 11.81
move_nanmax 19.18 6.43 19.70 22.74 15.41 27.09
Reference functions:
median np.median
nanmedian local copy of sp.stats.nanmedian
nansum np.nansum
nanmax np.nanmax
nanmean local copy of sp.stats.nanmean
nanstd local copy of sp.stats.nanstd
nanargmax np.nanargmax
ss scipy.stats.ss
rankdata scipy.stats.rankdata based (axis support added)
partsort np.sort, n=max(a.shape[0]/2,1)
argpartsort np.argsort, n=max(a.shape[0]/2,1)
move_sum sp.ndimage.convolve1d based, window=a.shape[0]/5
move_nansum sp.ndimage.convolve1d based, window=a.shape[0]/5
move_mean sp.ndimage.convolve1d based, window=a.shape[0]/5
move_nanmean sp.ndimage.convolve1d based, window=a.shape[0]/5
move_std sp.ndimage.convolve1d based, window=a.shape[0]/5
move_nanstd sp.ndimage.convolve1d based, window=a.shape[0]/5
move_max sp.ndimage.maximum_filter1d based, window=a.shape[0]/5
move_nanmax sp.ndimage.maximum_filter1d based, window=a.shape[0]/5Under the hood Bottleneck uses a separate Cython function for each combination of ndim, dtype, and axis. A lot of the overhead in bn.nanmax(), for example, is in checking that the axis is within range, converting non-array data to an array, and selecting the function to use to calculate the maximum.
You can get rid of the overhead by doing all this before you, say, enter an inner loop:
>>> arr = np.random.rand(10,10)
>>> func, a = bn.func.nansum_selector(arr, axis=0)
>>> func
<built-in function nansum_2d_float64_axis0> Let's see how much faster than runs:
>>> timeit np.nansum(arr, axis=0)
10000 loops, best of 3: 20.4 us per loop
>>> timeit bn.nansum(arr, axis=0)
100000 loops, best of 3: 2.05 us per loop
>>> timeit func(a)
100000 loops, best of 3: 1.14 us per loopNote that func is faster than Numpy's non-NaN version of sum:
>>> timeit arr.sum(axis=0)
100000 loops, best of 3: 3.03 us per loopSo, in this example, adding NaN protection to your inner loop comes at a negative cost!
Benchmarks for the low-level Cython functions:
>>> bn.bench(mode='faster', dtype='float64', axis=0)
Bottleneck performance benchmark
Bottleneck 0.5.0
Numpy (np) 1.6.0
Scipy (sp) 0.9.0
Speed is NumPy or SciPy time divided by Bottleneck time
NaN means one-third NaNs; float64 and axis=0 are used
Low-level functions used (mode='faster')
no NaN no NaN no NaN NaN NaN NaN
(10,10) (100,100) (1000,1000) (10,10) (100,100) (1000,1000)
median 6.99 2.14 2.23 8.11 4.57 2.95
nanmedian 155.70 29.55 4.25 199.93 73.12 6.48
nansum 16.06 5.14 1.69 16.17 6.07 1.68
nanmax 3.55 1.35 1.01 4.11 4.29 1.09
nanmean 37.91 10.77 2.37 38.41 26.50 4.41
nanstd 45.15 7.71 2.36 47.14 15.85 3.38
nanargmax 15.12 5.64 2.64 15.74 8.74 2.82
ss 8.46 2.72 1.21 8.34 2.71 1.22
rankdata 24.46 13.16 8.94 24.32 14.39 9.82
partsort 2.33 2.17 2.44 2.82 6.23 3.80
argpartsort 1.15 2.17 1.67 1.15 3.81 1.67
move_sum 16.89 8.80 14.17 16.76 9.18 13.80
move_nansum 42.53 21.27 28.58 44.57 26.91 29.30
move_mean 15.70 4.55 14.32 15.96 9.01 13.86
move_nanmean 49.32 11.71 29.30 50.52 14.58 30.32
move_std 22.64 3.45 22.96 32.23 22.24 29.30
move_nanstd 46.44 6.10 34.35 56.45 6.91 36.26
move_max 5.86 3.80 9.32 6.98 6.05 11.70
move_nanmax 26.30 6.38 19.52 32.37 15.48 27.06
Reference functions:
median np.median
nanmedian local copy of sp.stats.nanmedian
nansum np.nansum
nanmax np.nanmax
nanmean local copy of sp.stats.nanmean
nanstd local copy of sp.stats.nanstd
nanargmax np.nanargmax
ss scipy.stats.ss
rankdata scipy.stats.rankdata based (axis support added)
partsort np.sort, n=max(a.shape[0]/2,1)
argpartsort np.argsort, n=max(a.shape[0]/2,1)
move_sum sp.ndimage.convolve1d based, window=a.shape[0]/5
move_nansum sp.ndimage.convolve1d based, window=a.shape[0]/5
move_mean sp.ndimage.convolve1d based, window=a.shape[0]/5
move_nanmean sp.ndimage.convolve1d based, window=a.shape[0]/5
move_std sp.ndimage.convolve1d based, window=a.shape[0]/5
move_nanstd sp.ndimage.convolve1d based, window=a.shape[0]/5
move_max sp.ndimage.maximum_filter1d based, window=a.shape[0]/5
move_nanmax sp.ndimage.maximum_filter1d based, window=a.shape[0]/5Currently only 1d, 2d, and 3d input arrays with data type (dtype) int32, int64, float32, and float64 are accelerated. All other ndim/dtype combinations result in calls to slower, unaccelerated functions.
Bottleneck is distributed under a Simplified BSD license. Parts of NumPy, Scipy and numpydoc, all of which have BSD licenses, are included in Bottleneck. See the LICENSE file, which is distributed with Bottleneck, for details.
Requirements:
| Bottleneck | Python 2.5, 2.6, 2.7; NumPy 1.5.1 or 1.6.0 |
| Unit tests | nose |
| Compile | gcc or MinGW |
| Optional | SciPy 0.8.0 or 0.9.0 (portions of benchmark) |
Directions for installing a released version of Bottleneck (i.e., one obtained from http://pypi.python.org/pypi/Bottleneck) are given below. Cython is not required since the Cython files have already been converted to C source files. (If you obtained bottleneck directly from the repository, then you will need to generate the C source files using the included Makefile which requires Cython.)
GNU/Linux, Mac OS X, et al.
To install Bottleneck:
$ python setup.py build
$ sudo python setup.py installOr, if you wish to specify where Bottleneck is installed, for example inside /usr/local:
$ python setup.py build
$ sudo python setup.py install --prefix=/usr/localWindows
You can compile Bottleneck using the instructions below or you can use the Windows binaries created by Christoph Gohlke: http://www.lfd.uci.edu/~gohlke/pythonlibs/#bottleneck
In order to compile the C code in Bottleneck you need a Windows version of the gcc compiler. MinGW (Minimalist GNU for Windows) contains gcc.
Install MinGW and add it to your system path. Then install Bottleneck with the commands:
python setup.py build --compiler=mingw32
python setup.py installPost install
After you have installed Bottleneck, run the suite of unit tests:
>>> import bottleneck as bn
>>> bn.test()
<snip>
Ran 80 tests in 49.602s
OK
<nose.result.TextTestResult run=80 errors=0 failures=0>