gptf is a library for building Guassian Process models in Python using TensorFlow, based on GPflow. Its benefits over GPflow include:
- Ops can be easily pinned to devices / graphs, and inherit their device placement from their parents.
- Autoflow that plays nicely with the distributed runtime.
- Better trees for a better world.
Explanatory notebooks can be found in the notebooks directory, and documentation can be found here.
- Install TensorFlow manually (at least version 0.9).
- Run
python setup.py develop
.
Tests can be run using python setup.py nosetests
.