This is the implementation of the scalable heteroscedastic GP (HGP) developed in "Haitao Liu, Yew-Soon Ong, and Jianfei Cai, Large-scale Heteroscedastic Regression via Gaussian Process." Please see the paper for further details.
We here focus on the heteroscedastic Gaussian process regression
To improve the scalability of HGP, we introduce inducing variables and perform variational inference to derive an analytical evidence lower bound (ELBO), the part of which factorizes over data points, thus supporting stochastic variational inference for handling large-scale heteroscedastic regression.
The model is implemented based on GPflow. To use it, one can
- download GPflow package from here;
- use
likelihoods.py
to replace the same file./gpflow/likelihoods.py
in GPflow; - move
svhgp.py
to the folder./gpflow/models/
; - install GPflow through
python setup.py install
.
We have tested the model using GPflow 1.3.0.
The illustration example is provided in
demo_SVHGP_toy.ipynb