This repository implements an Interp1d class that overrides torch.autograd.Function, enabling
linear 1D interpolation on the GPU for Pytorch.
class Interp1
67C1
d(torch.autograd.Function):
def forward(ctx, 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 Matlab™ 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
-
x: a (N, ) or (D, N) Pytorch Tensor: Either 1-D or 2-D. It contains the coordinates of the observed samples. -
y: (N,) or (D, N) Pytorch Tensor. Either 1-D or 2-D. It contains the actual values that correspond to the coordinates given byx. The length ofyalong its last dimension must be the same as that ofx -
xnew: (P,) or (D, P) Pytorch Tensor. Either 1-D or 2-D. If it is not 1-D, its length along the first dimension must be the same as that of whicheverxandyis 2-D. x-coordinates for which we want the interpolated output. -
out: (D, P) Pytorch Tensor` Tensor for the output. If None: allocated automatically.
a Pytorch tensor of shape (D, P), containing the interpolated values.
The CUDA interp1d function depends on the torchsearchsorted repository.
If not installed, you must:
- Clone that repo through
git clone git@github.com:aliutkus/torchsearchsorted.git - Go in the corresponding directory of that repo and launch
python setup.py install.
Then, type pip install -e . in the root folder of this repo.
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%.