torchist 0.1.0

NumPy-style histograms in PyTorch

NumPy-style histograms in PyTorch

The torchist package implements NumPy's histogram and histogramdd functions in PyTorch with support for non-uniform binning. The package also features implementations of ravel_multi_index, unravel_index and some useful functionals (e.g. 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 using

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

or copy the package directly to your project, with

git clone https://github.com/francois-rozet/torchist
cp -R torchist/torchist <path/to/project>/torchist

Getting Started

import torch
import torchist

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

hist = torchist.histogramdd(x, bins=10, low=0., upp=1.)

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

Benchmark

The implementations of torchist are up to 3 times faster than those of numpy on CPU and benefit greately from CUDA capabilities.

$ python torchist/__init__.py
CPU
---
np.histogram : 1.3613 s
np.histogramdd : 19.8844 s
np.histogram (non-uniform) : 5.5652 s
np.histogramdd (non-uniform) : 17.5668 s
torchist.histogram : 0.9674 s
torchist.histogramdd : 6.3047 s
torchist.histogram (non-uniform) : 3.6520 s
torchist.histogramdd (non-uniform) : 14.1086 s

CUDA
----
torchist.histogram : 0.1032 s
torchist.histogramdd : 0.2668 s
torchist.histogram (non-uniform) : 0.1230 s
torchist.histogramdd (non-uniform) : 0.4407 s