NumPy-style histograms in PyTorch

The torchist package implements NumPy's histogram and histogramdd functions in PyTorch with CUDA support. The package also features implementations of ravel_multi_index, unravel_index and some useful functionals like entropy or kl_divergence.

Installation

The torchist package is available on PyPI, which means it is installable with pip.

pip install torchist

Alternatively, if you need the latest features, you can install it from the repository.

pip install git+https://github.com/francois-rozet/torchist

Getting Started

import torch
import torchist

x = torch.rand(100, 3).cuda()

hist = torchist.histogramdd(x, bins=10, low=0.0, upp=1.0)

print(hist.shape)  # (10, 10, 10)

Benchmark

The implementations of torchist are on par or faster than those of numpy and torch on CPU. On GPU (CUDA), they are much faster.

$ torchist-benchmark
CPU
---
np.histogram                      0.8917 s
np.histogram (edges)              0.5993 s
np.histogramdd                   16.8441 s
np.histogramdd (edges)           13.7680 s
torch.histogram                   0.3251 s
torch.histogram (edges)           0.4217 s
torch.histogramdd                 1.0528 s
torch.histogramdd (edges)         1.1955 s
torchist.histogram                0.4250 s
torchist.histogram (edges)        0.6372 s
torchist.histogramdd              1.6266 s
torchist.histogramdd (edges)      3.8619 s

CUDA
----
torchist.histogram                0.1045 s
torchist.histogram (edges)        0.0672 s
torchist.histogramdd              0.0906 s
torchist.histogramdd (edges)      0.1170 s