https://github.com/GPflow/GPflow

sort by:
Revision Author Date Message Commit Date
445112b Merge remote-tracking branch 'origin/GPflow-1.0-RC' into john-bradshaw/binary-class-GP 24 October 2017, 10:52 UTC
fcd8ff2 Notebook wording and typo improvements. 24 October 2017, 10:52 UTC
f96d0ed Assign operation for parameterized object. 24 October 2017, 10:09 UTC
4303206 Merge branch 'GPflow-1.0-RC' of github.com:GPflow/GPflow into GPflow-1.0-RC 24 October 2017, 09:00 UTC
0d7f6ae HMC wrap output in pandas DataFrame. 24 October 2017, 09:00 UTC
33a8d17 Add licence top at node.py 23 October 2017, 23:05 UTC
a7854ef HMC output as dictionary. 23 October 2017, 22:07 UTC
be3162f Hidden names for parameters and parameterized gpflow structures. 23 October 2017, 21:44 UTC
7790327 Merge branch 'GPflow-1.0-RC' of github.com:GPflow/GPflow into GPflow-1.0-RC 23 October 2017, 17:23 UTC
0d9e027 AutoFlow functions accept optional arguments for tf.Session().run 23 October 2017, 17:22 UTC
39a5410 improvements to notebook wording 23 October 2017, 16:22 UTC
d2f5017 Added a EP versus VGP Binary classification notebook to serve as some documentation. 23 October 2017, 16:16 UTC
e71fe63 Merge remote-tracking branch 'origin/GPflow-1.0-RC' into john-bradshaw/binary-class-GP # Conflicts: # gpflow/__init__.py 23 October 2017, 10:36 UTC
5b16a12 Refactor of EP implementation to modularise and make more extensible in the future. * EP while loop broken out and made into seperate class to allow it to be easy to reuse elsewhere. Tidy up of the variable names and made more general. Tests written to check that it can recover the true solution when the approximating factors have the same solution as the true solution. * Central moment functions moved out of EP class and live elsewhere. Step function, Gaussian and Gaussian CDF moments implemented all with tests. * Likelihoods now are associated with a method that returns their central moments generator function. Only implemented for Bernoulli but this allows one to store moment logic with the likelihood function which is more where it belongs. * Updated the VGP versus EP binary classification example script to use this new source structure. This has remained in the docs so have not written tests for this part. 23 October 2017, 10:32 UTC
3d07e9d Type casting at Parameter. 22 October 2017, 21:55 UTC
2a6673b Add dtype option to Parameter. 22 October 2017, 19:26 UTC
3d406c4 Use right objective function inside hmc. 21 October 2017, 21:45 UTC
0fa5e2c HMC - strict result unfolding in condition. 21 October 2017, 10:20 UTC
99658ed Fixes to extentions of tensor lists. 20 October 2017, 21:47 UTC
cb7c385 HMC trainable_tensors is a generator, not a list. 20 October 2017, 15:14 UTC
9a1834d HMC rename wrong property. 20 October 2017, 14:50 UTC
01a5567 Export only HMC optimizer. 19 October 2017, 23:42 UTC
1ecc551 HMC minimize method raises exception correctly. 19 October 2017, 21:50 UTC
ea36507 HMC bugfixing. 19 October 2017, 21:46 UTC
b0acd6f Finishing work on HMC. 19 October 2017, 15:43 UTC
d9a4eeb Clean up hmc mess. 18 October 2017, 23:08 UTC
13673d5 HMC updates. 17 October 2017, 22:06 UTC
de3ebe3 Detached objective function. Decentralize creation. 15 October 2017, 20:58 UTC
f46eeb3 Using `params_as_tensors_for(obj, convert=False)` instead of duplicated contextmanager to get temporary access to params following @awav suggestion. 10 October 2017, 13:33 UTC
9f2cf21 Merge remote-tracking branch 'origin/GPflow-1.0-RC' into john-bradshaw/binary-class-GP 10 October 2017, 10:54 UTC
77be468 Binary classification with EP * fixed a bug in log likelihood that was causing NaNs * fixed the gradients for likelihood as was doing too many tf.stop_gradients * added the option to not rerun EP each time you call log_likelihood * tidied up the demo so that get nice clean comparison between VGP and EP * soem numerical fixes and minor refactoring of EP code. 10 October 2017, 10:53 UTC
f7b52ab adding a stable way to compute deriv log cdf of a normal dist. 09 October 2017, 14:18 UTC
e534ceb Clear method for nodes didn't work correctly. 08 October 2017, 23:15 UTC
d81b3c0 Move HMC file. 08 October 2017, 16:17 UTC
12bd394 Minibatch tensor name is aligned with model name. Parameters as tensors with option. 08 October 2017, 16:17 UTC
f8f0811 Fix shape method of dataholders. 07 October 2017, 17:58 UTC
9946b8a Parentable class moved to separate module. 07 October 2017, 17:26 UTC
e63590f Re-initialization issue is gone, i.e. you can change values of parameter and compile it again, you will not see previous initialization value again, the parameter will be initialized with latest one. Uniquness of parentable object is no longer a requirement: new parameter or model object - new tensors in the graph with similar name. 07 October 2017, 16:52 UTC
ad6dcec EP -- binary classifications -- building up. Demo seens sensible. 06 October 2017, 15:51 UTC
0400eb8 Empty parameterized objects doesn't have priors. No parameters - priors is None. 05 October 2017, 21:35 UTC
c864b0c Proper minibatch size setting in new GPflow. 04 October 2017, 22:50 UTC
a5210f6 Advanced minibatching. Volatile batch size. 04 October 2017, 22:17 UTC
04ab8b3 Merge branch 'GPflow-1.0-RC' of github.com:GPflow/GPflow into GPflow-1.0-RC 04 October 2017, 14:19 UTC
0cc7db7 Parameter formatting fixed. Externally defined property available for every descendant. 04 October 2017, 14:19 UTC
7db2d8c improvements to coreg and classification demo 04 October 2017, 14:11 UTC
a8770a1 improvements to kernel notebook 04 October 2017, 13:57 UTC
d0c9c68 minor edit to the initialization of SVGP to allow for non-minibatching version 04 October 2017, 13:44 UTC
8f03ea0 Merge branch 'GPflow-1.0-RC' of github.com:GPflow/GPflow into GPflow-1.0-RC 04 October 2017, 12:53 UTC
8d5f246 modified notebooks for gpflow 1.0 (unfinished) 04 October 2017, 12:53 UTC
2393ae8 Parameter formatting fixed. Predict test cases fixed. 04 October 2017, 10:46 UTC
a5c01a8 Priors dimension fixed. 04 October 2017, 09:25 UTC
a67ba2c Bump version to 0.9.8. 03 October 2017, 23:01 UTC
8276398 Revert back classification.ipynb 03 October 2017, 22:51 UTC
314c5bd Model name option argument is returned. Constrained parameter for Bernoulli had wrong shape. 03 October 2017, 22:44 UTC
efcd60f Merge branch 'GPflow-1.0-RC' of github.com:GPflow/GPflow into GPflow-1.0-RC 03 October 2017, 19:15 UTC
2a71583 New project layout - add params sub module. 03 October 2017, 19:12 UTC
c47bc5c Rename Param to Parameter [INTERNALLY]. Minibatch has new implementation. 03 October 2017, 19:11 UTC
b9b4ed7 adding softplus as needed in transforms 03 October 2017, 17:10 UTC
d955dd8 Priors test cases fixed. 01 October 2017, 16:53 UTC
c3e6b37 Fixing likelihood's tests. 1 error left caused by old minibatch module. 30 September 2017, 17:58 UTC
439bb9c GPLVM tests are fixed. 29 September 2017, 20:01 UTC
c30478e Put in order unconstrained and constrained variables at codebase. 29 September 2017, 16:17 UTC
9617299 Kernel test fixes in progress. 28 September 2017, 21:30 UTC
cc5e7d0 Fixes in kernel testing. 28 September 2017, 14:49 UTC
d5c8585 Parameter accepts lists. Ekernels tests fixed. 27 September 2017, 23:03 UTC
73ef1c6 In progress of fixing kernel tests. Spotted an issue with initializing. 27 September 2017, 19:33 UTC
2ea1cf9 Progress on fixing ekernels tests. Clear method checks unnecessary build status. 27 September 2017, 18:37 UTC
6c5f503 Fixing old models tests. 26 September 2017, 23:21 UTC
e0a2cc7 Parameter list represents only parameters. Empty method returns true only if there are no data holders and parameters at parameterized object. 26 September 2017, 20:28 UTC
6381a83 None prior for parameterized object without parameters. 26 September 2017, 18:58 UTC
2d7f140 Address an issue with initializing AdamOptimizer. 26 September 2017, 18:36 UTC
ee53610 Transform is an named optional parameter, hence should be used appropriately at parameter creation. It caused error in SVGP model during optimization - kernel variance became negative, as transform wasn't setup properly and covariance matrix blow up with garbage covariance matrix. 26 September 2017, 14:39 UTC
d6883d2 Fixing model tests. 26 September 2017, 08:27 UTC
eb0c98c Params tests are passing. 25 September 2017, 16:30 UTC
baf110d Fix SVGP model. 24 September 2017, 22:58 UTC
635aa5f Autoflow fix. Add param tests. 24 September 2017, 19:26 UTC
b392b46 Param tests. 24 September 2017, 16:15 UTC
03364d3 Rename accesses to models and parameters in tests. 24 September 2017, 08:55 UTC
1e16309 Add test util. 24 September 2017, 07:59 UTC
582745f Testing context with graph and session. 24 September 2017, 07:59 UTC
a7063fc Reset model at TensorFlow optimizer. 23 September 2017, 20:38 UTC
cee7902 Remove print functions. 23 September 2017, 20:35 UTC
cb8ee93 Optimizers finalizing. 23 September 2017, 20:31 UTC
0f03240 Training part is finished. 23 September 2017, 16:14 UTC
795369b Add gpflow/core sub-folder. 22 September 2017, 20:38 UTC
6b305a9 Re-organizing code. Aliases for settings parameters. 22 September 2017, 20:37 UTC
eb1f0e5 Set -x mode for params.py 20 September 2017, 23:33 UTC
30074d7 Fixing setup.py file. Bump version. 20 September 2017, 17:25 UTC
c58d278 Create __init__ file in models folder. 20 September 2017, 10:00 UTC
4b41d91 Change models folder layout. 20 September 2017, 09:59 UTC
a20e233 External optimization skeleton. 19 September 2017, 20:57 UTC
5620d34 Add non-empty params list property. 19 September 2017, 20:56 UTC
4a82fa1 TensorConverter bug found and fixed. 19 September 2017, 16:53 UTC
6bd507a is_built method requires graph argument. 18 September 2017, 23:23 UTC
11a510f Code organization. 18 September 2017, 21:32 UTC
1153278 Check if external tensor is trainable before start working with it. 18 September 2017, 12:35 UTC
86e4223 The clean can be initiated by the node only if it wasn't built. 17 September 2017, 21:20 UTC
0629779 Pylint changes. DataHolder inherits main implementations from Param. 17 September 2017, 20:30 UTC
4e8ce45 Fixing autoflow 15 September 2017, 23:08 UTC
064243b Update pylintrc. 15 September 2017, 21:49 UTC
back to top