https://github.com/GPflow/GPflow
Revision 5641e9a618cd0ef58befbdb32b3d9b2d494563ee authored by John Bradshaw on 25 September 2017, 12:51:55 UTC, committed by John Bradshaw on 25 September 2017, 12:51:55 UTC
Many kernels correspond to finite feature mappings. Using these features is often useful as they scaled better as a function of number of data points (but badly in terms of sdcaling with feature dimension). * this commit introduces a feature mapping transform that can be obtained from some kernels. * for some kernels eg linear we implement the exact transform * the stationary kernels have feature approximations implemented using the random features approach of Rahmini & Recht * unit tests to check that these are working roughly correctly.
1 parent 75733ff
Tip revision: 5641e9a618cd0ef58befbdb32b3d9b2d494563ee authored by John Bradshaw on 25 September 2017, 12:51:55 UTC
Linearise kernels -- feature map transforms for kernels.
Linearise kernels -- feature map transforms for kernels.
Tip revision: 5641e9a
File | Mode | Size |
---|---|---|
doc | ||
gpflow | ||
notebooks | ||
testing | ||
.coveragerc | -rw-r--r-- | 261 bytes |
.coveralls.yml | -rw-r--r-- | 23 bytes |
.gitignore | -rw-r--r-- | 793 bytes |
.travis.yml | -rw-r--r-- | 842 bytes |
Dockerfile | -rw-r--r-- | 1.1 KB |
LICENSE | -rw-r--r-- | 11.1 KB |
MANIFEST.in | -rw-r--r-- | 166 bytes |
README.md | -rw-r--r-- | 5.0 KB |
RELEASE.md | -rw-r--r-- | 3.5 KB |
contributing.md | -rw-r--r-- | 4.1 KB |
docs_require.txt | -rw-r--r-- | 404 bytes |
roadmap.md | -rw-r--r-- | 506 bytes |
run_tests.sh | -rwxr-xr-x | 896 bytes |
setup.py | -rw-r--r-- | 2.5 KB |
Computing file changes ...