https://github.com/GPflow/GPflow
Revision 4d628f1716c24f9ef6552b57d4d021c101620ba1 authored by James Hensman on 28 June 2016, 18:53:15 UTC, committed by James Hensman on 28 June 2016, 18:53:15 UTC
Kernels with active dimensions now select columns of data using
tensorflow's gather, rather that a packed series of scalar indexes.

old: tf.pack([X[:,i] for in in active_dims])
new: tf.transpose(tf.gather(tf.transpose(X), active_dims))

the latter is preferred because the graph is much smaller, especially
for a large nuber of active dimensions.

In addition, the kernel tests have been tidied and extended to cover
more of kernels.py.
1 parent 1cb047c
History
Tip revision: 4d628f1716c24f9ef6552b57d4d021c101620ba1 authored by James Hensman on 28 June 2016, 18:53:15 UTC
Small improvements to kernel slicing
Tip revision: 4d628f1
File Mode Size
GPflow
notebooks
testing
.coveragerc -rw-r--r-- 251 bytes
.coveralls.yml -rw-r--r-- 23 bytes
.gitignore -rw-r--r-- 702 bytes
LICENSE -rw-r--r-- 11.1 KB
README.md -rw-r--r-- 5.0 KB
RELEASE.md -rw-r--r-- 1.2 KB
setup.py -rw-r--r-- 1.4 KB

README.md

back to top