Skip to content
master
Go to file
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 
 
 
 
 
 
 
 
 

README.rst

Numbagg: Fast N-dimensional aggregation functions with Numba

https://travis-ci.org/shoyer/numbagg.svg?branch=master

Fast, flexible N-dimensional array functions written with Numba and NumPy's generalized ufuncs.

Currently accelerated functions:

  • Array functions: allnan, anynan, count, nanargmax, nanargmin, nanmax, nanmean, nanstd, nanvar, nanmin, nansum
  • Moving window functions: move_exp_nanmean, move_mean

Note: Only functions listed here (exposed in Numbagg's top level namespace) are supported as part of Numbagg's public API.

Easy to extend

Numbagg makes it easy to write, in pure Python/NumPy, flexible aggregation functions accelerated by Numba. All the hard work is done by Numba's JIT compiler and NumPy's gufunc machinery (as wrapped by Numba).

For example, here is how we wrote nansum:

import numpy as np
from numbagg.decorators import ndreduce

@ndreduce
def nansum(a):
    asum = 0.0
    for ai in a.flat:
        if not np.isnan(ai):
            asum += ai
    return asum

You are welcome to experiment with Numbagg's decorator functions, but these are not public APIs (yet): we reserve the right to change them at any time.

We'd rather get your pull requests to add new functions into Numbagg directly!

Advantages over Bottleneck

  • Way less code. Easier to add new functions. No ad-hoc templating system. No Cython!
  • Fast functions still work for >3 dimensions.
  • axis argument handles tuples of integers.

Most of the functions in Numbagg (including our test suite) are adapted from Bottleneck's battle-hardened implementations. Still, Numbagg is experimental, and probably not yet ready for production.

Benchmarks

Initial benchmarks are quite encouraging. Numbagg/Numba has comparable (slightly better) performance than Bottleneck's hand-written C:

import numbagg
import numpy as np
import bottleneck

x = np.random.RandomState(42).randn(1000, 1000)
x[x < -1] = np.NaN

# timings with numba=0.41.0 and bottleneck=1.2.1

In [2]: %timeit numbagg.nanmean(x)
1.8 ms ± 92.3 µs per loop (mean ± std. dev. of 7 runs, 1 loop each)

In [3]: %timeit numbagg.nanmean(x, axis=0)
3.63 ms ± 136 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

In [4]: %timeit numbagg.nanmean(x, axis=1)
1.81 ms ± 41 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

In [5]: %timeit bottleneck.nanmean(x)
2.22 ms ± 119 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

In [6]: %timeit bottleneck.nanmean(x, axis=0)
4.45 ms ± 107 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

In [7]: %timeit bottleneck.nanmean(x, axis=1)
2.19 ms ± 13.7 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

Our approach

Numbagg includes somewhat awkward workarounds for features missing from NumPy/Numba:

I hope that the need for most of these will eventually go away. In the meantime, expect Numbagg to be tightly coupled to Numba and NumPy release cycles.

License

3-clause BSD. Includes portions of Bottleneck, which is distributed under a Simplified BSD license.

About

Fast N-dimensional aggregation functions with Numba

Topics

Resources

License

Languages

You can’t perform that action at this time.