require python=2.7
conda create -n py27 python=2.7
conda activate py27
pip install -r requirement.txt
python experiments/single_output.py
cd experiment
python xxx.py
This repository contains the implementation of our Continual (Multi-task) Gaussian Process model. We provide a detailed code for single-output GP regression, multi-output GP regression, GP classification and long-term continual learning.
Please, if you use this code, cite the following preprint:
@article{MorenoArtesAlvarez19,
title = {Continual Multi-task Gaussian Processes},
author = {Moreno-Mu\~noz, Pablo and Art\'es-Rodr\'iguez, Antonio and \'Alvarez, Mauricio A},
journal = {arXiv preprint arXiv:1911.00002},
year = {2019}
}
Solar sunspots data.
Results: In the /experiments/ folder you may find the following scripts for simulations.
single_output.py // Continual GP regression
multi_output.py // Continual multi-output GP regression
banana.py // Continual GP classification
solar.py // Long-term continual GP regression (figure above).
The Python syntaxes of likelihood distributions and the structure of our code is based on the HetMOGP repository.
Pablo Moreno-Muñoz, Antonio Artés-Rodríguez and Mauricio A. Álvarez
For further information or contact:
pmoreno@tsc.uc3m.es