Numbagg: Fast N-dimensional aggregation functions with Numba

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, nanquantile, ffill, bfill.
• Grouped functions: group_nanall, group_nanany, group_nanargmax, group_nanargmin, group_nancount, group_nanfirst, group_nanlast, group_nanmax, group_nanmean, group_nanmin, group_nanprod, group_nanstd, group_nansum, group_nansum_of_squares, group_nanvar.
• Moving window functions listed below
• Exponentially weighted moving functions listed below

Why use numba?

Performance

• Much faster than pandas for almost every function — 2-20x
• About the same speed as bottleneck on a single calculation
• Much faster than bottleneck — 4-7x — when parallelizing with multiple cores — for example, calculating over each row on an array with 10 rows.
• ...though numbagg's functions are JIT compiled, so they're much slower on their first run

Versatility

• More functions (though bottleneck has some functions we don't have, and pandas' functions have many more parameters)
• Fast functions work for >3 dimensions. Functions take an arbitrary axis or tuple of axes to calculate over
• Written in numba — way less code, simple to inspect, simple to improve

Benchmarks

2D

Array of shape (1, 10000000), over the final axis

ND

Array of shape (100, 1000, 1000), over the final axis

[^1][^2][^3]

[^1]: Benchmarks were run on a Mac M1 laptop in October 2023 on numbagg's HEAD, pandas 2.1.1, bottleneck 1.3.7. They're also run in CI, though without demonstrating the benefits of parallelization given GHA's CPU count.

[^2]: While we separate the setup and the running of the functions, pandas still needs to do some work to create its result dataframe, and numbagg does some checks in python which bottleneck does in C or doesn't do. So we focus on the benchmarks for larger arrays in order to reduce that impact. Any contributions to improve the benchmarks are welcome.

[^3]: Pandas doesn't have an equivalent move_exp_nancount function, so this is compared to a function which uses its sum function on an array of 1s.

Example implementation

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

Implementation details

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

• It implements its own cache for functions wrapped by Numba's guvectorize, because that decorator is rather slow.
• It does its own handling of array transposes to handle the axis argument in reduction functions.
• It rewrites plain functions into gufuncs, to allow writing a traditional function while retaining the multidimensional advantages of gufuncs.

Already some of the ideas here have flowed upstream to numba (for example, an axis parameter), and we hope that others will follow.

License

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