https://github.com/GPflow/GPflow
Revision 69364c2b79e1e6dd0fe7419377a044c0bdedb4f5 authored by Mark van der Wilk on 04 October 2017, 18:13:22 UTC, committed by GitHub on 04 October 2017, 18:13:22 UTC
* Clean up kullback_leiblers: a single function for whitened/unwhitened and diagonal/full-covariance. Deprecate the rest. * Update VGP and SVGP models to use new `gauss_kl` * clean up test_conditionals * Improve documentation * tidy up gauss_kl code * Add dedicated tests for Gaussian KL divergence * Bumped version. * Added encoding hint for python2.7. * @ to .dot() * change encoding to utf-8 * Added coding for `test_kldiv.py` + PEP8. * I'm all for high test coverage, but this seems a bit overzealous :P
1 parent 4d8b552
Tip revision: 69364c2b79e1e6dd0fe7419377a044c0bdedb4f5 authored by Mark van der Wilk on 04 October 2017, 18:13:22 UTC
Clean up kullback_leiblers (#514)
Clean up kullback_leiblers (#514)
Tip revision: 69364c2
roadmap.md
This document covers major planned development items for GPflow.
# Computational speed
- Add further benchmarks to [benchmark repository](https://github.com/GPflow/GPflowBenchmarks) including multiple GPUs.
- Incorporate remaining Tom Nickson GPU code from [branch](https://github.com/c0g/tomserflow) into TensorFlow main.
# Features
- Add ability to exploit Kronecker structure.
# Housekeeping
- See also issues marked "enhancement" in the GitHub [list](https://github.com/GPflow/GPflow/issues).
Computing file changes ...