Interactive Gaussian Processes

The collection of GP visualizations. The GP computation part benefits from TensorFlow 2.0 and GPflow 2.0, whereas the visualization implemetation sits on top of the holoviews framework, which in turn based on bokeh.

Any contributions or ideas about visualizations which you think could be helpful are welcome.

Implemented:

• GP samples for squared exponential, matern52, matern32, matern12, arccosine 0th order and linear kernels

Plans to add:

• GPR model visualization with different kernels and ability to add new data points
• SVGP online training, moving inducing points

Install package

Using conda is optional, but still it is recommended in the case when you don't want to collide with existing packages like TensorFlow or GPflow

conda activate your-env
(your-env) pip install -r requirements.txt
(your-env) pip install -e .

How to run interactive HTML page:

(your-env) bokeh serve --show apps/samples.py

You can run the same plots in a notebook - check notebooks folder.