Bin Stat with CUDA Acceleration

Introduction

Implementing statistical tools that I commonly use in Python

• np.histogram
• np.histogram2d
• scipy.stats.binned_statistic
• scipy.stats.binned_statistic2d

The code is written in CUDA/C++ and ported to Python with pybind11.

Install

pip install -e . 

Usage

import numpy as np
import binstatcuda as bsc

n_sample = int(1e7)
x_edges = np.linspace(0.0, 1.0, 128, dtype=np.float32)
y_edges = np.linspace(0.0, 1.0, 64, dtype=np.float32)

x = np.random.random(n_sample).astype(np.float32)
y = np.random.random(n_sample).astype(np.float32)

hist_1d = bsc.histogram(x, bins=x_edges)
hist_2d = bsc.histogram2d(x, y, bins=(x_edges, y_edges))

values = np.random.random(n_sample).astype(np.float32)

binned_mean = bsc.binned_statistic(
    x=x,
    values=values,
    statistic='mean',
    bins=x_edges,
)

binned_mean_2d = bsc.binned_statistic_2d(
    x=x,
    y=y,
    values=values,
    statistic='mean',
    bins=(x_edges, y_edges)
)

Supported statistic methods are,

• mean
• std
• sum
• count

Check demo/demo.html for more examples.

Note

• all float point arithematic are performed in float, inplicit casting happens
• all counting are performed in unsigned long long

TODO

Implement more statistical methods for binned_statistic and binned_statistic_2d:

• min
• max
• median

Implement more functions

• np.histogramdd
• scipy.stats.binned_statisticdd