Tools for exact GP inference on massive image, video, spatial-temporal, or multi-output datasets.
Only python 2 is supported currently, and up-to-date Anaconda distribution is recommended.
For basic usage, we only rely on the numpy and scipy packages.
Optionally, the lightweight tabulate
package can be installed with pip
which we use for printing.
Additional non-standard packages such as GPy
are required for testing, however, this is not required for regular use.
- Gaussian Process Inference on Full Grids: This simple tutorial takes you though performing efficient Gaussian process inference on fully structure grids with no missing observations.
- Gaussian Process Inference on Grids with Missing Observations: This tutorial considers the general case where the input data is structured on a grid, however, some input response are missing. It will consider multi-output senarios as well as senarios where several input dimensions form a dimension of a grid.
The underlying algorithms are based on the 2018 SDM paper (found here):
@inproceedings{evans_gp_grid,
title={Exploiting Structure for Fast Kernel Learning},
author={Evans, Trefor W and Nair, Prasanth B},
booktitle={SIAM International Conference on Data Mining},
year={2018},
pages={414-422}
}