Skip to content
Code for Randomly Projected Additive Gaussian Processes
Python Jupyter Notebook Shell
Branch: master
Clone or download
Latest commit 8e2876e Jan 13, 2020
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
fitting
gp_models partial fix for gpytorch 1.0 compatibility Jan 12, 2020
model_specs implemented mixed exact gp with optimization and model averaging Jul 29, 2019
run_scripts add GAM run script Jul 18, 2019
.gitignore
Examples.ipynb adding Examples notebook Nov 20, 2019
README.md
config_template.py
gp_experiment_runner.py Add a more info about data paths Jan 12, 2020
rp.py
synthetic_test_script.py updated synthetic test script Aug 12, 2019
test.py finish cleanup, tests working Aug 7, 2019
training_routines.py merge changes keeping double/float conversion and recording warnings Aug 13, 2019
utils.py first, mostly untested cut at Compressed GP sampling Jul 29, 2019

README.md

Randomly Projected Additive GPs git repo

This repo contains implementations and experiment code for the paper Randomly Projected Additive Gaussian Processes for Regression

Requirements

  • Python > 3.0
  • GPyTorch >= 1.0
  • PyKeOps >= 1.2

Files

  • config_template.py: Template configuration file for dataset file locations, etc. Rename to config.py and replace with your file configurations. UCI datasets referenced in the experiments may be downloaded here.
  • gp_experiment_runner.py: Command-line endpoint used for running batches of experiments.
  • synthetic_test_script.py: A simple script for running synthetic experiments.
  • rp.py: Generating (random) projection matrices, including a routine for generating diversified projection matrices (useed in DPA-GP).
  • training_routines.py: A collection of routines used to construct, train, and test GPs in this project.
  • test.py: a suite of unit tests.
  • utils.py: Utilities that are reused and don't live in a particular section of the project.

Subpackages

  • gp_models: Encapsulates the model (and kernel) definitions for kernels and models used.
  • fitting: Encapsulates methods for learning. Currently, only optimization-based methods are available, as opposed to, e.g., sampling.

Folders

  • model_specs: Model specification .json files. These are used to store and re-use the configuration of models.
  • run_scripts: Re-used/example command-line calls to gp_experiment_runner.py.

UCI Data Sets

To download the UCI data sets used for benchmarks, download them from Andrew Gordon Wilson's home page. See config_template.py for details on how these files are expected to be organized in accordance with your configurations.

You can’t perform that action at this time.