Skip to content
No description, website, or topics provided.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Decoupled Gaussian process model.

This is not an officially supported Google product.

This repository contains an implementation of the Decoupled Gaussian Process model that decouples the representation of mean and covariance in reproducing kernel Hilbert space.

The details of the model is in the paper: Cheng, Ching-An, and Byron Boots. "Variational Inference for Gaussian Process Models with Linear Complexity." Advances in Neural Information Processing Systems. 2017.

Link to the paper:

How to use the model

This model can be used mainly in two ways:

  • through (detailed in, and
  • through tf.estimator.Estimator with model_fn() (defined in


File provides detailed steps to train and evaluate the model by first building the graph, and then iteratively minimizing the objective function by In order to use the model as a layer, you may want to embed the logic of adding bases online and hyperparameters initialization in the graph, so that no initial values for hyperparameters are needed and no need to call model.bases.add_bases() in the train loop anymore.

Through model_fn()

model_fn() is defined in We can use the following code to create an tf.estimator.Estimator:

estimator = tf.estimator.Estimator(

Then the tf.estimator.Estimator can be used with tf.contrib.learn.Experiment, to quickly carry out experiments to compare against other models.

You can’t perform that action at this time.