Home

Awesome

torchinterp1d

CUDA 1-D interpolation for Pytorch

Requires PyTorch >= 1.6 (due to torch.searchsorted).

Presentation

This repository implements an interp1d function that overrides torch.autograd.Function, enabling linear 1D interpolation on the GPU for Pytorch.

def interp1d(x, y, xnew, out=None)

This function returns interpolated values of a set of 1-D functions at the desired query points xnew.

It works similarly to Matlabâ„¢ or scipy functions with the linear interpolation mode on, except that it parallelises over any number of desired interpolation problems and exploits CUDA on the GPU

Parameters for interp1d

Results

a Pytorch tensor of shape (D, P), containing the interpolated values.

Installation

Type pip install -e . in the root folder of this repo.

Usage

Basically simply calle torchinterp1d.interp1d.

Try out python test.py in the examples folder.

Solving 100000 interpolation problems: each with 100 observations and 30 desired values
CPU: 8060.260ms, GPU: 70.735ms, error: 0.000000%.