Skip to content


Subversion checkout URL

You can clone with
Download ZIP
Fast NumPy array functions written in Cython
Python C Makefile
Fetching latest commit...
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.




Bottleneck is a collection of fast, NumPy array functions written in Cython.

The three categories of Bottleneck functions:

  • Faster replacement for NumPy and SciPy functions
  • Moving window functions
  • Group functions that bin calculations by like-labeled elements

Function signatures (using nanmean as an example):

Functions nanmean(arr, axis=None)
Moving window move_mean(arr, window, axis=0)
Group by group_nanmean(arr, label, order=None, axis=0)

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)

Moving window nanmean:

>>> bn.move_nanmean(arr, window=2)
array([ nan,  1.5,  2. ,  4. ,  4.5])

Group nanmean:

>>> label = ['a', 'a', 'b', 'b', 'a']
>>> bn.group_nanmean(arr, label)
(array([ 2.66666667,  4.        ]), ['a', 'b'])


Bottleneck is fast:

>>> arr = np.random.rand(100, 100)
>>> timeit np.nanmax(arr)
10000 loops, best of 3: 99.6 us per loop
>>> timeit bn.nanmax(arr)
100000 loops, best of 3: 15.3 us per loop

Let's not forget to add some NaNs:

>>> arr[arr > 0.5] = np.nan
>>> timeit np.nanmax(arr)
10000 loops, best of 3: 146 us per loop
>>> timeit bn.nanmax(arr)
100000 loops, best of 3: 15.2 us per loop

Bottleneck comes with a benchmark suite that compares the performance of the bottleneck functions that have a NumPy/SciPy equivalent. To run the benchmark:

>>> bn.benchit(verbose=False)
Bottleneck performance benchmark
    Bottleneck  0.1.0dev
    Numpy       1.5.1
    Scipy       0.8.0
    Speed is numpy (or scipy) time divided by Bottleneck time
    NaN means all NaNs
   Speed   Test                  Shape        dtype    NaN?
   2.4019  median(a, axis=-1)    (500,500)    float64
   2.2668  median(a, axis=-1)    (500,500)    float64  NaN
   4.1235  median(a, axis=-1)    (10000,)     float64
   4.3498  median(a, axis=-1)    (10000,)     float64  NaN
   9.8184  nanmax(a, axis=-1)    (500,500)    float64
   7.9157  nanmax(a, axis=-1)    (500,500)    float64  NaN
   9.2306  nanmax(a, axis=-1)    (10000,)     float64
   8.1635  nanmax(a, axis=-1)    (10000,)     float64  NaN
   6.7218  nanmin(a, axis=-1)    (500,500)    float64
   7.9112  nanmin(a, axis=-1)    (500,500)    float64  NaN
   6.4950  nanmin(a, axis=-1)    (10000,)     float64
   8.0791  nanmin(a, axis=-1)    (10000,)     float64  NaN
  12.3650  nanmean(a, axis=-1)   (500,500)    float64
  42.0738  nanmean(a, axis=-1)   (500,500)    float64  NaN
  12.2769  nanmean(a, axis=-1)   (10000,)     float64
  22.1285  nanmean(a, axis=-1)   (10000,)     float64  NaN
   9.5515  nanstd(a, axis=-1)    (500,500)    float64
  68.9192  nanstd(a, axis=-1)    (500,500)    float64  NaN
   9.2174  nanstd(a, axis=-1)    (10000,)     float64
  26.1753  nanstd(a, axis=-1)    (10000,)     float64  NaN


Under 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.nanmax_selector(arr, axis=0)
>>> func
<built-in function nanmax_2d_float64_axis0>

Let's see how much faster than runs:

>> timeit np.nanmax(arr, axis=0)
10000 loops, best of 3: 25.7 us per loop
>> timeit bn.nanmax(arr, axis=0)
100000 loops, best of 3: 5.25 us per loop
>> timeit func(a)
100000 loops, best of 3: 2.5 us per loop

Note that func is faster than Numpy's non-NaN version of max:

>> timeit arr.max(axis=0)
100000 loops, best of 3: 3.28 us per loop

So adding NaN protection to your inner loops comes at a negative cost!


Bottleneck is in the prototype stage.

Bottleneck contains the following functions:

nanmean move_nanmean group_nanmean

Currently only 1d, 2d, and 3d NumPy arrays with dtype int32, int64, and float64 are supported.


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.


mailing list



Bottleneck Python, NumPy 1.5.1+, SciPy 0.8.0+
Unit tests nose
Compile gcc or MinGW

GNU/Linux, Mac OS X, et al.

To install Bottleneck:

$ python build
$ sudo python install

Or, if you wish to specify where Bottleneck is installed, for example inside /usr/local:

$ python build
$ sudo python install --prefix=/usr/local


In order to compile the C code in dsna you need a Windows version of the gcc compiler. MinGW (Minimalist GNU for Windows) contains gcc and has been used to successfully compile dsna on Windows.

Install MinGW and add it to your system path. Then install dsna with the commands:

python build --compiler=mingw32
python install

Post install

After you have installed Bottleneck, run the suite of unit tests:

>>> import bottleneck as bn
>>> bn.test()
Ran 10 tests in 13.756s
<nose.result.TextTestResult run=10 errors=0 failures=0>
Something went wrong with that request. Please try again.