{"author":null,"author_email":"Stephan Hoyer , Maximilian Roos ","bugtrack_url":null,"classifiers":["Intended Audience :: Science/Research","Operating System :: OS Independent","Programming Language :: Python","Programming Language :: Python :: 3","Programming Language :: Python :: 3 :: Only","Programming Language :: Python :: 3.10","Programming Language :: Python :: 3.11","Programming Language :: Python :: 3.12","Programming Language :: Python :: 3.9","Topic :: Scientific/Engineering"],"description":"# Numbagg: Fast N-dimensional aggregation functions with Numba\n\n[![GitHub Workflow CI Status](https://img.shields.io/github/actions/workflow/status/numbagg/numbagg/test.yaml?branch=main&logo=github&style=for-the-badge)](https://github.com/numbagg/numbagg/actions/workflows/test.yaml)\n[![PyPI Version](https://img.shields.io/pypi/v/numbagg?style=for-the-badge)](https://pypi.python.org/pypi/numbagg/)\n\nFast, flexible N-dimensional array functions written with\n[Numba](https://github.com/numba/numba) and NumPy's [generalized\nufuncs](http://docs.scipy.org/doc/numpy/reference/c-api.generalized-ufuncs.html).\n\n## Why use numbagg?\n\n### Performance\n\n- Outperforms pandas\n - On a single core, 2-10x faster for moving window functions, 1-2x faster for\n aggregation and grouping functions\n - When parallelizing with multiple cores, 4-30x faster\n- Outperforms bottleneck on multiple cores\n - On a single core, matches bottleneck\n - When parallelizing with multiple cores, 3-7x faster\n- Outperforms numpy on multiple cores\n - On a single core, matches numpy\n - When parallelizing with multiple cores, 5-15x faster\n- ...though numbagg's functions are JIT compiled, so the first run is much slower\n\n### Versatility\n\n- More functions (though bottleneck has some functions we don't have, and pandas' functions\n have many more parameters)\n- Functions work for >3 dimensions. All functions take an arbitrary axis or\n tuple of axes to calculate over\n- Written in numba — way less code, simple to inspect, simple to improve\n\n## Functions & benchmarks\n\n### Summary benchmark\n\nTwo benchmarks summarize numbagg's performance — the first with a 1D array of 10M elements without\nparallelization, and a second with a 2D array of 100x10K elements with parallelization. Numbagg's relative\nperformance is much higher where parallelization is possible. A wider range of arrays is\nlisted in the full set of benchmarks below.\n\nThe values in the table are numbagg's performance as a multiple of other libraries for a\ngiven shaped array calculated over the final axis. (so 1.00x means numbagg is equal,\nhigher means numbagg is faster.)\n\n| func | 1D
pandas | 1D
bottleneck | 1D
numpy | 2D
pandas | 2D
bottleneck | 2D
numpy |\n| :------------------------ | -----------: | ---------------: | ----------: | -----------: | ---------------: | ----------: |\n| `bfill` | 1.06x | 1.13x | n/a | 11.11x | 5.04x | n/a |\n| `ffill` | 1.12x | 0.99x | n/a | 11.50x | 4.25x | n/a |\n| `group_nanall` | 1.38x | n/a | n/a | 7.77x | n/a | n/a |\n| `group_nanany` | 1.12x | n/a | n/a | 6.21x | n/a | n/a |\n| `group_nanargmax` | 1.16x | n/a | n/a | 6.81x | n/a | n/a |\n| `group_nanargmin` | 1.17x | n/a | n/a | 6.48x | n/a | n/a |\n| `group_nancount` | 1.05x | n/a | n/a | 4.94x | n/a | n/a |\n| `group_nanfirst` | 1.52x | n/a | n/a | 11.13x | n/a | n/a |\n| `group_nanlast` | 1.12x | n/a | n/a | 5.56x | n/a | n/a |\n| `group_nanmax` | 1.13x | n/a | n/a | 5.13x | n/a | n/a |\n| `group_nanmean` | 1.14x | n/a | n/a | 5.61x | n/a | n/a |\n| `group_nanmin` | 1.12x | n/a | n/a | 5.75x | n/a | n/a |\n| `group_nanprod` | 1.15x | n/a | n/a | 5.25x | n/a | n/a |\n| `group_nanstd` | 1.14x | n/a | n/a | 5.41x | n/a | n/a |\n| `group_nansum_of_squares` | 1.33x | n/a | n/a | 8.00x | n/a | n/a |\n| `group_nansum` | 1.18x | n/a | n/a | 5.63x | n/a | n/a |\n| `group_nanvar` | 1.13x | n/a | n/a | 4.88x | n/a | n/a |\n| `move_corr` | 16.42x | n/a | n/a | 115.76x | n/a | n/a |\n| `move_cov` | 12.30x | n/a | n/a | 86.56x | n/a | n/a |\n| `move_exp_nancorr` | 6.65x | n/a | n/a | 46.98x | n/a | n/a |\n| `move_exp_nancount` | 1.88x | n/a | n/a | 9.95x | n/a | n/a |\n| `move_exp_nancov` | 6.53x | n/a | n/a | 43.63x | n/a | n/a |\n| `move_exp_nanmean` | 1.61x | n/a | n/a | 10.65x | n/a | n/a |\n| `move_exp_nanstd` | 1.76x | n/a | n/a | 12.40x | n/a | n/a |\n| `move_exp_nansum` | 1.09x | n/a | n/a | 9.01x | n/a | n/a |\n| `move_exp_nanvar` | 1.77x | n/a | n/a | 11.41x | n/a | n/a |\n| `move_mean` | 6.03x | 1.34x | n/a | 26.60x | 6.25x | n/a |\n| `move_std` | 4.76x | 0.89x | n/a | 29.09x | 6.24x | n/a |\n| `move_sum` | 5.16x | 1.13x | n/a | 24.02x | 6.10x | n/a |\n| `move_var` | 5.45x | 1.05x | n/a | 29.54x | 6.05x | n/a |\n| `nanargmax`[^5] | 2.40x | 0.53x | n/a | 2.32x | 0.93x | n/a |\n| `nanargmin`[^5] | 2.35x | 0.50x | n/a | 2.53x | 1.00x | n/a |\n| `nancount` | 2.01x | n/a | 1.59x | 12.26x | n/a | 3.96x |\n| `nanmax`[^5] | 3.15x | 0.50x | 0.09x | 3.59x | 3.24x | 0.09x |\n| `nanmean` | 3.00x | 1.01x | 3.82x | 18.98x | 5.04x | 19.33x |\n| `nanmin`[^5] | 3.07x | 0.50x | 0.09x | 3.39x | 3.03x | 0.09x |\n| `nanquantile` | 0.69x | n/a | 0.53x | 4.94x | n/a | 4.33x |\n| `nanstd` | 1.63x | 1.61x | 3.39x | 12.39x | 10.18x | 22.03x |\n| `nansum` | 2.48x | 0.94x | 3.31x | 20.47x | 4.65x | 17.90x |\n| `nanvar` | 1.61x | 1.65x | 3.40x | 12.62x | 10.49x | 22.13x |\n\n### Full benchmarks\n\n

\n\n| func | shape | size | ndim | pandas | bottleneck | numpy | numbagg | pandas_ratio | bottleneck_ratio | numpy_ratio | numbagg_ratio |\n| :------------------------ | ---------------------: | --------: | ---: | -----: | ---------: | -----: | ------: | -----------: | ---------------: | ----------: | ------------: |\n| `bfill` | (1000,) | 1000 | 1 | 0ms | 0ms | n/a | 0ms | 0.38x | 0.01x | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 15ms | 16ms | n/a | 14ms | 1.06x | 1.13x | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 37ms | 17ms | n/a | 3ms | 11.11x | 5.04x | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | 18ms | n/a | 3ms | n/a | 6.13x | n/a | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | 199ms | n/a | 31ms | n/a | 6.44x | n/a | 1.00x |\n| `ffill` | (1000,) | 1000 | 1 | 0ms | 0ms | n/a | 0ms | 0.37x | 0.01x | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 15ms | 14ms | n/a | 14ms | 1.12x | 0.99x | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 37ms | 14ms | n/a | 3ms | 11.50x | 4.25x | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | 14ms | n/a | 3ms | n/a | 4.64x | n/a | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | 176ms | n/a | 31ms | n/a | 5.72x | n/a | 1.00x |\n| `group_nanall` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.72x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 48ms | n/a | n/a | 35ms | 1.38x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 18ms | n/a | n/a | 2ms | 7.77x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 1ms | n/a | n/a | n/a | 1.00x |\n| `group_nanany` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.70x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 49ms | n/a | n/a | 44ms | 1.12x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 18ms | n/a | n/a | 3ms | 6.21x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 2ms | n/a | n/a | n/a | 1.00x |\n| `group_nanargmax` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 1.07x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 49ms | n/a | n/a | 42ms | 1.16x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 17ms | n/a | n/a | 3ms | 6.81x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 2ms | n/a | n/a | n/a | 1.00x |\n| `group_nanargmin` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 1.06x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 49ms | n/a | n/a | 42ms | 1.17x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 17ms | n/a | n/a | 3ms | 6.48x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 2ms | n/a | n/a | n/a | 1.00x |\n| `group_nancount` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.66x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 44ms | n/a | n/a | 42ms | 1.05x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 13ms | n/a | n/a | 3ms | 4.94x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 1ms | n/a | n/a | n/a | 1.00x |\n| `group_nanfirst` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.73x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 52ms | n/a | n/a | 34ms | 1.52x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 16ms | n/a | n/a | 1ms | 11.13x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 1ms | n/a | n/a | n/a | 1.00x |\n| `group_nanlast` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.72x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 47ms | n/a | n/a | 42ms | 1.12x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 14ms | n/a | n/a | 2ms | 5.56x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 1ms | n/a | n/a | n/a | 1.00x |\n| `group_nanmax` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.71x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 48ms | n/a | n/a | 43ms | 1.13x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 14ms | n/a | n/a | 3ms | 5.13x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 2ms | n/a | n/a | n/a | 1.00x |\n| `group_nanmean` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.72x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 50ms | n/a | n/a | 44ms | 1.14x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 16ms | n/a | n/a | 3ms | 5.61x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 2ms | n/a | n/a | n/a | 1.00x |\n| `group_nanmin` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.73x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 48ms | n/a | n/a | 43ms | 1.12x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 14ms | n/a | n/a | 2ms | 5.75x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 2ms | n/a | n/a | n/a | 1.00x |\n| `group_nanprod` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.70x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 48ms | n/a | n/a | 42ms | 1.15x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 14ms | n/a | n/a | 3ms | 5.25x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 1ms | n/a | n/a | n/a | 1.00x |\n| `group_nanstd` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.71x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 51ms | n/a | n/a | 45ms | 1.14x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 17ms | n/a | n/a | 3ms | 5.41x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 2ms | n/a | n/a | n/a | 1.00x |\n| `group_nansum` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.74x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 51ms | n/a | n/a | 43ms | 1.18x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 16ms | n/a | n/a | 3ms | 5.63x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 2ms | n/a | n/a | n/a | 1.00x |\n| `group_nanvar` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.70x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 51ms | n/a | n/a | 45ms | 1.13x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 16ms | n/a | n/a | 3ms | 4.88x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 2ms | n/a | n/a | n/a | 1.00x |\n| `group_nansum_of_squares` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.88x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 57ms | n/a | n/a | 43ms | 1.33x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 22ms | n/a | n/a | 3ms | 8.00x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 1ms | n/a | n/a | n/a | 1.00x |\n| `move_corr` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 2.68x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 710ms | n/a | n/a | 43ms | 16.42x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 683ms | n/a | n/a | 6ms | 115.76x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 5ms | n/a | n/a | n/a | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | n/a | n/a | 49ms | n/a | n/a | n/a | 1.00x |\n| `move_cov` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 2.43x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 490ms | n/a | n/a | 40ms | 12.30x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 460ms | n/a | n/a | 5ms | 86.56x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 4ms | n/a | n/a | n/a | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | n/a | n/a | 44ms | n/a | n/a | n/a | 1.00x |\n| `move_mean` | (1000,) | 1000 | 1 | 0ms | 0ms | n/a | 0ms | 0.46x | 0.01x | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 92ms | 21ms | n/a | 15ms | 6.03x | 1.34x | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 88ms | 21ms | n/a | 3ms | 26.60x | 6.25x | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | 20ms | n/a | 3ms | n/a | 6.66x | n/a | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | 228ms | n/a | 32ms | n/a | 7.12x | n/a | 1.00x |\n| `move_std` | (1000,) | 1000 | 1 | 0ms | 0ms | n/a | 0ms | 0.53x | 0.02x | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 141ms | 26ms | n/a | 30ms | 4.76x | 0.89x | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 123ms | 26ms | n/a | 4ms | 29.09x | 6.24x | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | 26ms | n/a | 4ms | n/a | 7.37x | n/a | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | 291ms | n/a | 37ms | n/a | 7.82x | n/a | 1.00x |\n| `move_sum` | (1000,) | 1000 | 1 | 0ms | 0ms | n/a | 0ms | 0.46x | 0.01x | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 95ms | 21ms | n/a | 18ms | 5.16x | 1.13x | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 83ms | 21ms | n/a | 3ms | 24.02x | 6.10x | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | 21ms | n/a | 3ms | n/a | 6.79x | n/a | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | 227ms | n/a | 31ms | n/a | 7.29x | n/a | 1.00x |\n| `move_var` | (1000,) | 1000 | 1 | 0ms | 0ms | n/a | 0ms | 0.50x | 0.02x | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 131ms | 25ms | n/a | 24ms | 5.45x | 1.05x | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 122ms | 25ms | n/a | 4ms | 29.54x | 6.05x | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | 25ms | n/a | 4ms | n/a | 7.12x | n/a | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | 275ms | n/a | 36ms | n/a | 7.69x | n/a | 1.00x |\n| `move_exp_nancorr` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 2.33x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 344ms | n/a | n/a | 52ms | 6.65x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 338ms | n/a | n/a | 7ms | 46.98x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 6ms | n/a | n/a | n/a | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | n/a | n/a | 55ms | n/a | n/a | n/a | 1.00x |\n| `move_exp_nancount` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.57x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 51ms | n/a | n/a | 27ms | 1.88x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 47ms | n/a | n/a | 5ms | 9.95x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 4ms | n/a | n/a | n/a | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | n/a | n/a | 40ms | n/a | n/a | n/a | 1.00x |\n| `move_exp_nancov` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 2.19x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 215ms | n/a | n/a | 33ms | 6.53x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 234ms | n/a | n/a | 5ms | 43.63x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 5ms | n/a | n/a | n/a | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | n/a | n/a | 43ms | n/a | n/a | n/a | 1.00x |\n| `move_exp_nanmean` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.39x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 47ms | n/a | n/a | 30ms | 1.61x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 52ms | n/a | n/a | 5ms | 10.65x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 4ms | n/a | n/a | n/a | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | n/a | n/a | 43ms | n/a | n/a | n/a | 1.00x |\n| `move_exp_nanstd` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.68x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 64ms | n/a | n/a | 36ms | 1.76x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 74ms | n/a | n/a | 6ms | 12.40x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 5ms | n/a | n/a | n/a | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | n/a | n/a | 44ms | n/a | n/a | n/a | 1.00x |\n| `move_exp_nansum` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.38x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 36ms | n/a | n/a | 33ms | 1.09x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 43ms | n/a | n/a | 5ms | 9.01x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 4ms | n/a | n/a | n/a | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | n/a | n/a | 42ms | n/a | n/a | n/a | 1.00x |\n| `move_exp_nanvar` | (1000,) | 1000 | 1 | 0ms | n/a | n/a | 0ms | 0.40x | n/a | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 56ms | n/a | n/a | 32ms | 1.77x | n/a | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 64ms | n/a | n/a | 6ms | 11.41x | n/a | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | n/a | 4ms | n/a | n/a | n/a | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | n/a | n/a | 46ms | n/a | n/a | n/a | 1.00x |\n| `nanargmax`[^5] | (1000,) | 1000 | 1 | 0ms | 0ms | n/a | 0ms | 17.65x | 0.17x | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 24ms | 5ms | n/a | 10ms | 2.40x | 0.53x | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 25ms | 10ms | n/a | 11ms | 2.32x | 0.93x | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | 11ms | n/a | 11ms | n/a | 1.00x | n/a | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | 107ms | n/a | 108ms | n/a | 0.99x | n/a | 1.00x |\n| `nanargmin`[^5] | (1000,) | 1000 | 1 | 0ms | 0ms | n/a | 0ms | 17.72x | 0.17x | n/a | 1.00x |\n| | (10000000,) | 10000000 | 1 | 25ms | 5ms | n/a | 11ms | 2.35x | 0.50x | n/a | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 25ms | 10ms | n/a | 10ms | 2.53x | 1.00x | n/a | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | 11ms | n/a | 11ms | n/a | 1.00x | n/a | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | 108ms | n/a | 108ms | n/a | 1.00x | n/a | 1.00x |\n| `nancount` | (1000,) | 1000 | 1 | 0ms | n/a | 0ms | 0ms | 0.77x | n/a | 0.02x | 1.00x |\n| | (10000000,) | 10000000 | 1 | 3ms | n/a | 3ms | 2ms | 2.01x | n/a | 1.59x | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 8ms | n/a | 3ms | 1ms | 12.26x | n/a | 3.96x | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | 3ms | 1ms | n/a | n/a | 3.97x | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | n/a | 33ms | 7ms | n/a | n/a | 5.07x | 1.00x |\n| `nanmax`[^5] | (1000,) | 1000 | 1 | 0ms | 0ms | 0ms | 0ms | 11.07x | 0.17x | 0.55x | 1.00x |\n| | (10000000,) | 10000000 | 1 | 32ms | 5ms | 1ms | 10ms | 3.15x | 0.50x | 0.09x | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 36ms | 33ms | 1ms | 10ms | 3.59x | 3.24x | 0.09x | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | 32ms | 1ms | 10ms | n/a | 3.24x | 0.10x | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | 320ms | 11ms | 98ms | n/a | 3.26x | 0.11x | 1.00x |\n| `nanmean` | (1000,) | 1000 | 1 | 0ms | 0ms | 0ms | 0ms | 0.39x | 0.00x | 0.05x | 1.00x |\n| | (10000000,) | 10000000 | 1 | 17ms | 6ms | 21ms | 6ms | 3.00x | 1.01x | 3.82x | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 21ms | 5ms | 21ms | 1ms | 18.98x | 5.04x | 19.33x | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | 5ms | 21ms | 1ms | n/a | 6.10x | 23.77x | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | 54ms | 258ms | 8ms | n/a | 7.00x | 33.59x | 1.00x |\n| `nanmin`[^5] | (1000,) | 1000 | 1 | 0ms | 0ms | 0ms | 0ms | 10.86x | 0.17x | 0.55x | 1.00x |\n| | (10000000,) | 10000000 | 1 | 33ms | 5ms | 1ms | 11ms | 3.07x | 0.50x | 0.09x | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 36ms | 32ms | 1ms | 11ms | 3.39x | 3.03x | 0.09x | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | 32ms | 1ms | 10ms | n/a | 3.12x | 0.10x | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | 320ms | 11ms | 102ms | n/a | 3.12x | 0.11x | 1.00x |\n| `nanquantile` | (1000,) | 1000 | 1 | 0ms | n/a | 0ms | 0ms | 0.56x | n/a | 0.21x | 1.00x |\n| | (10000000,) | 10000000 | 1 | 114ms | n/a | 87ms | 164ms | 0.69x | n/a | 0.53x | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 131ms | n/a | 115ms | 27ms | 4.94x | n/a | 4.33x | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | n/a | 315ms | 19ms | n/a | n/a | 16.51x | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | n/a | 3118ms | 165ms | n/a | n/a | 18.88x | 1.00x |\n| `nanstd` | (1000,) | 1000 | 1 | 0ms | 0ms | 0ms | 0ms | 0.31x | 0.02x | 0.14x | 1.00x |\n| | (10000000,) | 10000000 | 1 | 21ms | 20ms | 43ms | 13ms | 1.63x | 1.61x | 3.39x | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 24ms | 20ms | 43ms | 2ms | 12.39x | 10.18x | 22.03x | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | 20ms | 46ms | 1ms | n/a | 14.17x | 32.66x | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | 202ms | 513ms | 13ms | n/a | 16.08x | 40.78x | 1.00x |\n| `nansum` | (1000,) | 1000 | 1 | 0ms | 0ms | 0ms | 0ms | 0.46x | 0.01x | 0.03x | 1.00x |\n| | (10000000,) | 10000000 | 1 | 14ms | 5ms | 19ms | 6ms | 2.48x | 0.94x | 3.31x | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 22ms | 5ms | 19ms | 1ms | 20.47x | 4.65x | 17.90x | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | 5ms | 20ms | 1ms | n/a | 6.21x | 22.95x | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | 53ms | 226ms | 8ms | n/a | 6.98x | 29.90x | 1.00x |\n| `nanvar` | (1000,) | 1000 | 1 | 0ms | 0ms | 0ms | 0ms | 0.32x | 0.02x | 0.13x | 1.00x |\n| | (10000000,) | 10000000 | 1 | 21ms | 21ms | 44ms | 13ms | 1.61x | 1.65x | 3.40x | 1.00x |\n| | (100, 100000) | 10000000 | 2 | 25ms | 21ms | 43ms | 2ms | 12.62x | 10.49x | 22.13x | 1.00x |\n| | (10, 10, 10, 10, 1000) | 10000000 | 5 | n/a | 20ms | 46ms | 1ms | n/a | 14.02x | 32.28x | 1.00x |\n| | (100, 1000, 1000) | 100000000 | 3 | n/a | 202ms | 503ms | 13ms | n/a | 15.68x | 38.98x | 1.00x |\n\n

\n\n[^1][^2][^3][^4]\n\n[^1]:\n Benchmarks were run on a Mac M3 Max laptop in September 2024 on numbagg's HEAD,\n pandas 2.2.2, bottleneck 1.4.0 numpy 2.0.1, with `python\nnumbagg/test/run_benchmarks.py -- --benchmark-max-time=10`. They run in CI,\n though GHA's low CPU count means we don't see the full benefits of\n parallelization.\n\n[^2]:\n While we separate the setup and the running of the functions, pandas still\n needs to do some work to create its result dataframe, and numbagg does some\n checks in python which bottleneck does in C or doesn't do. So use benchmarks\n on larger arrays for our summary so we can focus on the computational speed,\n which doesn't asymptote away. Any contributions to improve the benchmarks are\n welcome.\n\n[^3]:\n In some instances, a library won't have the exact function — for example,\n pandas doesn't have an equivalent `move_exp_nancount` function, so we use\n its `sum` function on an array of `1`s. Similarly for\n `group_nansum_of_squares`, we use two separate operations.\n\n[^4]:\n `anynan` & `allnan` are also functions in numbagg, but not listed here as they\n require a different benchmark setup.\n\n[^5]:\n This function is not currently parallelized, so exhibits worse performance\n on parallelizable arrays.\n\n## Example implementation\n\nNumbagg makes it easy to write, in pure Python/NumPy, flexible aggregation\nfunctions accelerated by Numba. All the hard work is done by Numba's JIT\ncompiler and NumPy's gufunc machinery (as wrapped by Numba).\n\nFor example, here is how we wrote `nansum`:\n\n```python\nimport numpy as np\nfrom numbagg.decorators import ndreduce\n\n@ndreduce.wrap()\ndef nansum(a):\n asum = 0.0\n for ai in a.flat:\n if not np.isnan(ai):\n asum += ai\n return asum\n```\n\n## Implementation details\n\nNumbagg includes somewhat awkward workarounds for features missing from\nNumPy/Numba:\n\n- It implements its own cache for functions wrapped by Numba's\n `guvectorize`, because that decorator is rather slow.\n- It does its [own handling of array\n transposes](https://github.com/numbagg/numbagg/blob/e166adae94b3be35497dcdc22772026df75af253/numbagg/decorators.py#L170-L181)\n to handle the `axis` argument in reduction functions.\n- It [rewrites plain functions into\n gufuncs](https://github.com/numbagg/numbagg/blob/e166adae94b3be35497dcdc22772026df75af253/numbagg/transform.py),\n to allow writing a traditional function while retaining the multidimensional advantages of\n gufuncs.\n\nAlready some of the ideas here have flowed upstream to numba (for example, [an\naxis parameter](https://github.com/numpy/numpy/issues/5197)), and we hope\nthat others will follow.\n\n## License\n\n3-clause BSD. Includes portions of Bottleneck, which is distributed under a\nSimplified BSD license.\n","description_content_type":"text/markdown","docs_url":null,"download_url":null,"downloads":{"last_day":-1,"last_month":-1,"last_week":-1},"dynamic":null,"home_page":null,"keywords":null,"license":"License\n =======\n \n Numbagg is distributed under a BSD 3-clause license. It includes portions of\n Bottleneck, which is distributed under a Simplified BSD license.\n \n Numbagg license\n ===============\n \n BSD 3-Clause License\n \n Copyright (c) 2008-2012, AQR Capital Management, LLC, Lambda Foundry, Inc. and PyData Development Team\n Copyright (c) 2014-2018, Stephan Hoyer and Numbagg contributors\n All rights reserved.\n \n Redistribution and use in source and binary forms, with or without\n modification, are permitted provided that the following conditions are met:\n \n * Redistributions of source code must retain the above copyright notice, this\n list of conditions and the following disclaimer.\n \n * Redistributions in binary form must reproduce the above copyright notice,\n this list of conditions and the following disclaimer in the documentation\n and/or other materials provided with the distribution.\n \n * Neither the name of the copyright holder nor the names of its\n contributors may be used to endorse or promote products derived from\n this software without specific prior written permission.\n \n THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS \"AS IS\"\n AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE\n IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE\n DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE\n FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL\n DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR\n SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER\n CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,\n OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE\n OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.\n \n Bottleneck license\n ==================\n \n Copyright (c) 2010-2014 Berkeley Analytics, LLC.\n All rights reserved.\n \n Redistribution and use in source and binary forms, with or without\n modification, are permitted provided that the following conditions are met:\n \n * Redistributions of source code must retain the above copyright notice,\n this list of conditions and the following disclaimer.\n \n * Redistributions in binary form must reproduce the above copyright\n notice, this list of conditions and the following disclaimer in the\n documentation and/or other materials provided with the distribution.\n \n THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS \"AS IS\"\n AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE\n IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE\n ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE\n LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR\n CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF\n SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS\n INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN\n CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)\n ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE\n POSSIBILITY OF SUCH DAMAGE.\n ","license_expression":null,"license_files":null,"maintainer":null,"maintainer_email":null,"name":"numbagg","package_url":"https://pypi.org/project/numbagg/","platform":null,"project_url":"https://pypi.org/project/numbagg/","project_urls":{"homepage":"https://github.com/numbagg/numbagg"},"provides_extra":["dev"],"release_url":"https://pypi.org/project/numbagg/0.9.0/","requires_dist":["numpy","numba","bottleneck; extra == \"dev\"","hypothesis; extra == \"dev\"","jq; platform_system != \"Windows\" and extra == \"dev\"","mypy; extra == \"dev\"","pandas; extra == \"dev\"","pre-commit; extra == \"dev\"","pytest-benchmark; extra == \"dev\"","pytest; extra == \"dev\"","ruff; extra == \"dev\"","setuptools_scm; extra == \"dev\"","tabulate; extra == \"dev\""],"requires_python":">=3.9","summary":"Fast N-dimensional aggregation functions with Numba","version":"0.9.0","yanked":false,"yanked_reason":null}

Dependency	Optional
numba	false
numpy	false
bottleneck	true
hypothesis	true
jq	true
mypy	true
pandas	true
pre-commit	true
pytest	true
pytest-benchmark	true
ruff	true
setuptools_scm	true
tabulate	true

Package search

numbagg

project_urls

Dependencies

All package metadata

Releases