7b59483 | Mark van der Wilk | 30 October 2019, 13:27:20 UTC | Fix set trainable method (#1124) | 30 October 2019, 13:27:20 UTC |
d2ad129 | Mark van der Wilk | 30 October 2019, 11:04:30 UTC | Fix training function, which refers to model in outside scope. (#1123) | 30 October 2019, 11:04:30 UTC |
b1e7b32 | Mark van der Wilk | 29 October 2019, 11:25:50 UTC | Removed warnings due to weird escape characters. (#1122) | 29 October 2019, 11:25:50 UTC |
a6f3043 | Artem Artemev | 27 October 2019, 17:21:33 UTC | Merge branch 'develop-2.0' of github.com:GPflow/GPflow into develop-2.0 | 27 October 2019, 17:21:33 UTC |
5d6336f | Artem Artemev | 27 October 2019, 16:41:50 UTC | GPflow 2.0 notebook update: SGPR upper bound (#1116) This PR updates the understanding/upper_bound.ipynb to work in gpflow2. It fixes the upper_bound code in gpflow/models/sgpr.py and restores the original (tighter) test tolerance that had both been changed by commit 45efebc The plots don't quite match up to what was in the GPflow1 version (nor does the tight bound in the 1-inducing-point example at the end), but otherwise this seems to work. | 27 October 2019, 16:46:49 UTC |
9b5bf76 | st-- | 27 October 2019, 16:41:50 UTC | GPflow 2.0 notebook update: SGPR upper bound (#1116) | 27 October 2019, 16:41:50 UTC |
f650c7b | st-- | 27 October 2019, 16:31:31 UTC | GPflow 2.0 notebook update: understanding/models.ipynb (#1118) | 27 October 2019, 16:31:31 UTC |
d048736 | st-- | 27 October 2019, 16:28:49 UTC | Change GPflow 2.0 scipy default method to L-BFGS-B (as in GPflow 1.0) (#1112) Explicitly sets the default optimization method for Scipy() to L-BFGS-B (as in gpflow1), instead of letting scipy.minimize() figure it out. This resulted in BFGS, which may be too aggressive when encountering complex optimization surfaces, so L-BFGS-B may be a more sensible default. | 27 October 2019, 16:28:49 UTC |
ca40b13 | Artem Artemev | 25 October 2019, 09:34:12 UTC | Natural Gradient for GPflow 2.0 (#1109) | 25 October 2019, 09:34:12 UTC |
b6cc953 | Eric Hammy | 23 October 2019, 15:21:17 UTC | Notebook fixed - broke in earlier PR (#1115) | 23 October 2019, 15:21:17 UTC |
cd3559f | Sergio Pascual | 23 October 2019, 12:30:08 UTC | Set summary format is now fixed (#1114) * Some changes in convolutional notebook * Default summary format now loaded in `print_summary` instead of tabulate_module_summary * Default summary format now loaded in `print_summary` instead of tabulate_module_summary * Default summary format now loaded in `print_summary` instead of tabulate_module_summary | 23 October 2019, 12:30:08 UTC |
330074a | st-- | 23 October 2019, 10:16:51 UTC | GPflow 2.0 notebook update: FITC vs VFE (#1104) * WIP * make it run * rerun notebook * rerun notebook after merging in bugfix * Call step_callback through scipy.minimize's callback procedure (at the end of each step); previous behaviour was to call step_callback at each evaluation of objective function * adjust callback signature * changes to make it work under gpflow2: method=L-BFGS-B, tol=1e-11. increase callback frequency * rerun notebook * remove from notebook test blacklist * fix Scipy | 23 October 2019, 10:16:51 UTC |
6e43fec | st-- | 22 October 2019, 16:54:27 UTC | Change step_callback behaviour of Scipy optimizer in gpflow2 to behave as in gpflow1 (#1111) Changes behaviour of step_callback so that the callback only gets called once per optimisation step, not once per objective function evaluation. Makes gpflow2's Scipy() behave more like gpflow1's ScipyOptimizer()/monitoring. * Call step_callback through scipy.minimize's callback procedure (at the end of each step); previous behaviour was to call step_callback at each evaluation of objective function * split off from unpack_tensors (now stateless) the new assign_tensors (assigns to variables) * change callback_func to pass variables and their current values to step_callback, not assigning/computing loss/gradients * fix type signature * Improve docstring | 22 October 2019, 16:54:27 UTC |
a186ca5 | Artem Artemev | 22 October 2019, 16:38:10 UTC | Traverse function for DAG (tf.Module) structures (#1108) * Rewrite get_leaf_components and deepcopy_components with traverse Write/update tests for both * Tape should compute gradients outside context * Explanation of the test * Some changes in convolutional notebook * Update tests/test_deepcopy.py Co-Authored-By: st-- <st--@users.noreply.github.com> * Reorganising recursion on traverse * Update tests/test_deepcopy.py Co-Authored-By: st-- <st--@users.noreply.github.com> * Update tests/test_deepcopy.py Co-Authored-By: st-- <st--@users.noreply.github.com> * Update gpflow/utilities/utilities.py Co-Authored-By: Eric Hammy <6815729+condnsdmatters@users.noreply.github.com> * Update gpflow/utilities/utilities.py Co-Authored-By: Eric Hammy <6815729+condnsdmatters@users.noreply.github.com> * Update gpflow/utilities/utilities.py Co-Authored-By: Eric Hammy <6815729+condnsdmatters@users.noreply.github.com> * Update gpflow/utilities/utilities.py Co-Authored-By: Eric Hammy <6815729+condnsdmatters@users.noreply.github.com> * Update gpflow/utilities/utilities.py Co-Authored-By: Eric Hammy <6815729+condnsdmatters@users.noreply.github.com> * Update gpflow/utilities/utilities.py Co-Authored-By: Eric Hammy <6815729+condnsdmatters@users.noreply.github.com> * Update gpflow/utilities/utilities.py Co-Authored-By: Eric Hammy <6815729+condnsdmatters@users.noreply.github.com> * Update gpflow/utilities/utilities.py Co-Authored-By: Eric Hammy <6815729+condnsdmatters@users.noreply.github.com> * Separating deepcopy_components from clear_cache_bijectors * Remove ListWrapper and _DictWrapper Co-Authored-By: Eric Hammy <6815729+condnsdmatters@users.noreply.github.com> | 22 October 2019, 16:38:10 UTC |
51b2465 | Artem Artemev | 22 October 2019, 16:32:18 UTC | [GPflow2.0] notebook update: sanity_check.ipynb (#1078) | 22 October 2019, 16:32:18 UTC |
a813515 | Felix Leibfried | 22 October 2019, 13:13:55 UTC | GPflow2.0 notebook update: coregionalisation.ipynb (#1072) * explicitly set method="L-BFGS-B" and rerun | 22 October 2019, 13:13:55 UTC |
b0ddc6d | st-- | 21 October 2019, 14:19:32 UTC | Notebook for Variational Fourier Features (#1062) Adds a notebook that demonstrates how to add inter-domain inducing variables to GPflow without having to touch the core code base, at the example of variational Fourier features (Hensman, Durrande, Solin, JMLR 18 (2018)). | 21 October 2019, 14:19:32 UTC |
33ded03 | Eric Hammy | 18 October 2019, 15:06:01 UTC | Ensure all tf_probability distributions take the correct dtype parameters (#1103) | 18 October 2019, 15:06:01 UTC |
deb4508 | Eric Hammy | 17 October 2019, 14:46:42 UTC | Fix hidden bug in SGPR (#1106) 1. Fix hidden bug in SGPR 2. Add the sgpr.compute_qu method from gpflow1 1. [Bug]. SGPR likelihoods were previously using full rank matrices instead of diagonal ones in both upper bound and likelihood calculation. Ie `Kdiag` was not "diag". This error was being masked by the intentional deactivation of tests comparing to the SGPR to the GPR, and what appears to be a hack to make tests working on the upper bound case. 2. [Migration]. Fixing the above broke another test, originally used for sgpr.compute_qu. The method sgpr.compute_qu had not been migrated from gpflow1, and a test that was meant to check it had been patched up to pass, erroneously. After speaking to @markvdw, concluded this method is useful, in particular to compare to SVGP model. The test has been patched up and the method ported to gpflow2. | 17 October 2019, 14:46:42 UTC |
3b2a2ee | Eric Hammy | 16 October 2019, 09:39:04 UTC | Change gpflow.base.Module attributes `parameter` and `trainable parameter` to be wrapped in a tuple, not generators (#1102) This matches tf2 behaviour [here](https://github.com/tensorflow/tensorflow/blob/1cf0898dd4331baf93fe77205550f2c2e6c90ee5/tensorflow/python/module/module.py#L143) and prevents accidentally trying to reuse the same the generator once exhausted. | 16 October 2019, 09:39:04 UTC |
c62ad93 | st-- | 15 October 2019, 10:08:20 UTC | GPflow 2.0 notebook update: Multiclass classification (#1092) | 15 October 2019, 10:08:20 UTC |
b474291 | st-- | 15 October 2019, 10:07:27 UTC | GPflow2.0 notebook update: kernel design (#1087) | 15 October 2019, 10:07:27 UTC |
4aef40d | st-- | 14 October 2019, 21:36:48 UTC | fix bug in log_prior when a Parameter has a transform but no prior (#1099) Fixes a bug in the optimization objective for all `BayesianModel` subclasses when the model has parameters that have a `transform` (e.g. `positive()`, for kernel lengthscales and variances) but no `prior` (i.e. when doing Maximum Likelihood Estimation (MLE) inference). The objective should be independent of the parametrisation of the parameters, i.e. `log_prior()` of the parameters should be zero. | 14 October 2019, 21:36:48 UTC |
c5b91dd | Eric Hammy | 11 October 2019, 14:56:18 UTC | Merge pull request #1098 from GPflow/st/minor_cleanup Minor cleanup | 11 October 2019, 14:56:18 UTC |
f4688ba | st-- | 11 October 2019, 13:44:17 UTC | Apply suggestions from code review Fix types Co-Authored-By: Eric Hammy <6815729+condnsdmatters@users.noreply.github.com> | 11 October 2019, 13:44:17 UTC |
0ee881e | Eric Hammy | 11 October 2019, 11:12:17 UTC | Merge pull request #1097 from GPflow/st/quickfix_regression_notebook_test Quickfix: enable notebook test of regression.ipynb | 11 October 2019, 11:12:17 UTC |
a57569c | ST John | 11 October 2019, 10:59:22 UTC | Merge branch 'develop-2.0' of github.com:GPflow/GPflow into st/minor_cleanup | 11 October 2019, 10:59:22 UTC |
5cef76e | ST John | 11 October 2019, 10:44:47 UTC | regression.ipynb is ready for testing but had not been removed from blacklist | 11 October 2019, 10:44:47 UTC |
0b368ce | st-- | 11 October 2019, 10:37:44 UTC | Fix behaviour of likelihood quadrature tests (#1083) | 11 October 2019, 10:37:44 UTC |
7e6028c | Felix Leibfried | 11 October 2019, 09:21:27 UTC | GPflow2.0 notebook update: varying_noise.ipynb (#1050) | 11 October 2019, 09:21:27 UTC |
ba28c56 | jordigraumo | 11 October 2019, 09:02:35 UTC | GPflow2.0 notebook update: regression.ipynb (#1076) * regression notebook update to gpflow2 * now it has the correct text * restart&run all, and some minor text changes * minor updates * Update regression.ipynb | 11 October 2019, 09:02:35 UTC |
8e2e5cf | st-- | 10 October 2019, 18:40:38 UTC | Remove tf_graphs_and_sessions notebook (#1096) | 10 October 2019, 18:40:38 UTC |
99e8988 | st-- | 09 October 2019, 20:19:05 UTC | Sync notebooks with develop (includes moving files from notebooks2 back to doc/source/notebooks) (#1073) | 09 October 2019, 20:19:05 UTC |
3a21b65 | ST John | 09 October 2019, 13:42:44 UTC | adjust import for py3.8 | 09 October 2019, 13:42:44 UTC |
d8bbeb7 | ST John | 09 October 2019, 13:42:34 UTC | make docstrings r"" | 09 October 2019, 13:42:34 UTC |
a8f9348 | ST John | 09 October 2019, 13:32:07 UTC | minor changes | 09 October 2019, 13:32:11 UTC |
760bf04 | ST John | 09 October 2019, 13:30:58 UTC | add noise_variance argument to GPR and SGPR | 09 October 2019, 13:30:58 UTC |
4dba79d | st-- | 08 October 2019, 10:24:15 UTC | Clean up dependencies and setup instructions (#1074) | 08 October 2019, 10:24:15 UTC |
5a5c34e | Sergio Pascual | 07 October 2019, 14:59:11 UTC | Merge pull request #1085 from GPflow/sergio_pasc/gpflow-2.0/updated-print-summary-tests Test on print_summary now checks string output straight | 07 October 2019, 14:59:11 UTC |
5654336 | Sergio Diaz | 07 October 2019, 14:50:13 UTC | merging | 07 October 2019, 14:50:13 UTC |
ecd0b82 | Sergio Diaz | 07 October 2019, 14:47:23 UTC | Updated method names and signatures | 07 October 2019, 14:47:23 UTC |
ab73c4b | Sergio Pascual | 07 October 2019, 14:31:46 UTC | Update gpflow/utilities/utilities.py Co-Authored-By: st-- <st--@users.noreply.github.com> | 07 October 2019, 14:31:46 UTC |
74cf61b | Sergio Diaz | 07 October 2019, 14:10:45 UTC | Test on print_summary now checks string output straight | 07 October 2019, 14:10:45 UTC |
cba9732 | st-- | 07 October 2019, 14:09:57 UTC | Remove default argument of the set_trainable (was False) (#1060) | 07 October 2019, 14:09:57 UTC |
e3c3f96 | James Hensman | 07 October 2019, 13:17:17 UTC | Fix bug in print_summary for composition kernels (#1071) Closes #1066 | 07 October 2019, 13:17:17 UTC |
2974b94 | st-- | 07 October 2019, 12:47:36 UTC | Improve gpflow2 code coverage by adding tests (#1079) * add tests of gpflow.config * make config tests pass * use raise NotImplementedError in abstractmethod so codecov does not count it * swap order of tests so that it ends up in the original state * pca_reduce: turn assert into ValueError * add tests for gplvm constructor checks * add pragma no cover to scipy optimizer type check * more NotImplementedErrors * update codecov.yml * add missing logdensities tests * Change code coverage threshold to 95% so that the build status reports actually mean something again * improve test_dtype_errorcheck * pragma: no cover for ValueError check * . | 07 October 2019, 12:47:36 UTC |
48933ab | st-- | 04 October 2019, 15:12:44 UTC | minor cleanup of RobustMax (#1075) | 04 October 2019, 15:12:44 UTC |
81ae0be | Artem Artemev | 02 October 2019, 20:22:57 UTC | Remove Natural Gradient to increase code coverage | 02 October 2019, 20:22:57 UTC |
36c4736 | James Hensman | 02 October 2019, 09:50:28 UTC | Fix test coregion (#1067) | 02 October 2019, 09:50:28 UTC |
57f176c | Artem Artemev | 02 October 2019, 09:31:38 UTC | Merge branch 'awav/gpflow-2.0' of github.com:GPflow/GPflow into awav/gpflow-2.0 | 02 October 2019, 09:31:38 UTC |
8dfba0c | Artem Artemev | 02 October 2019, 09:31:32 UTC | Remove hmc.py optimizer | 02 October 2019, 09:31:32 UTC |
6e57626 | st-- | 01 October 2019, 19:13:50 UTC | GPflow 2.0 notebook update: GPs for big data (#1048) | 01 October 2019, 19:13:50 UTC |
705380d | st-- | 01 October 2019, 15:39:35 UTC | Rename inducing_variables -> inducing_variable as model attribute (#1063) | 01 October 2019, 15:39:35 UTC |
a39317b | st-- | 01 October 2019, 15:36:05 UTC | GPflow 2.0 notebook update: classification (#1044) | 01 October 2019, 15:36:05 UTC |
f9c3052 | Mark van der Wilk | 25 September 2019, 16:31:29 UTC | Improved convolutional GP notebook. (#1061) | 25 September 2019, 16:31:29 UTC |
13dd839 | Mark van der Wilk | 25 September 2019, 12:17:01 UTC | Added convolutional code (#1034) | 25 September 2019, 12:17:01 UTC |
1aff010 | Mark van der Wilk | 24 September 2019, 09:53:14 UTC | Mark/gpflow 2.0/kl dispatch (#1035) * Multiple dispatch for KL divergences. * prevent overwriting dispatcher * fix SVGP | 24 September 2019, 09:53:14 UTC |
f359d17 | Mark van der Wilk | 24 September 2019, 08:13:41 UTC | Added more elaborate documentation to multioutput notebook. (#1054) | 24 September 2019, 08:13:41 UTC |
102bf48 | Artem Artemev | 22 September 2019, 15:43:11 UTC | Remove tf.function(autograph=True) in gplvm notebook | 22 September 2019, 15:43:11 UTC |
292e446 | Artem Artemev | 22 September 2019, 15:16:00 UTC | Wrap gplvm training in tf.function | 22 September 2019, 15:16:00 UTC |
db6d2fb | Artem Artemev | 22 September 2019, 14:57:52 UTC | Add ci_utils.py module to the notebooks | 22 September 2019, 14:57:52 UTC |
8c54b52 | Artem Artemev | 22 September 2019, 14:00:32 UTC | Import tensorflow in gplvm.ipynb | 22 September 2019, 14:00:32 UTC |
f35fd17 | Artem Artemev | 22 September 2019, 13:58:11 UTC | Update README.md | 22 September 2019, 13:58:11 UTC |
f63857b | Artem Artemev | 21 September 2019, 21:08:09 UTC | Temporary change in README.md file | 21 September 2019, 21:08:09 UTC |
41dce7d | Artem Artemev | 16 September 2019, 15:43:28 UTC | GPflow 2.0 notebook update: GPLVM | 16 September 2019, 15:43:28 UTC |
e1cb9e7 | Sergio Pascual | 16 September 2019, 10:40:55 UTC | GPflow 2.0 notebook update: Ordinal Regression (#1049) | 16 September 2019, 10:40:55 UTC |
aaee8b1 | Vincent Dutordoir | 16 September 2019, 10:33:47 UTC | GPflow2.0 notebook update: multioutput (#1047) | 16 September 2019, 10:33:47 UTC |
da5c727 | Nicolas Durrande | 16 September 2019, 10:31:57 UTC | GPflow 2.0 notebook update: kernels (#1046) | 16 September 2019, 10:31:57 UTC |
9a278ac | Artem Artemev | 16 September 2019, 10:18:39 UTC | Update notebook test | 16 September 2019, 10:18:39 UTC |
6e624ce | Artem Artemev | 16 September 2019, 10:14:33 UTC | Notebook tests | 16 September 2019, 10:14:33 UTC |
333ee64 | Artem Artemev | 15 September 2019, 21:25:22 UTC | Notebook test fix | 15 September 2019, 21:25:22 UTC |
abf245b | Artem Artemev | 15 September 2019, 21:18:57 UTC | Add notebooks to tests | 15 September 2019, 21:18:57 UTC |
8ae58f2 | Artem Artemev | 13 September 2019, 15:59:46 UTC | GPLVM tests | 13 September 2019, 15:59:46 UTC |
708e383 | Artem Artemev | 13 September 2019, 12:58:16 UTC | Merge changes from develop | 13 September 2019, 12:58:16 UTC |
6dc68be | Artem Artemev | 13 September 2019, 12:40:04 UTC | Copy notebooks | 13 September 2019, 12:40:04 UTC |
a21a668 | Artem Artemev | 13 September 2019, 10:17:36 UTC | Merge branch 'sergio_pasc/gpflow-2.0/move-coregion-tests' into awav/gpflow-2.0 | 13 September 2019, 10:17:36 UTC |
09464b5 | Artem Artemev | 13 September 2019, 08:51:04 UTC | Update GPflow 2.0 notebook | 13 September 2019, 08:51:04 UTC |
3b7423f | Sergio Diaz | 12 September 2019, 12:30:56 UTC | Method equivalence and VGP, SGPR and SGPFITC fixed | 12 September 2019, 12:30:56 UTC |
45efebc | Sergio Diaz | 12 September 2019, 12:19:43 UTC | Method equivalence and VGP, SGPR and SGPFITC fixed | 12 September 2019, 12:19:43 UTC |
f174d7b | Sergio Diaz | 11 September 2019, 13:02:02 UTC | Tests for equivalence methods still in unsorted but updated to TF2 | 11 September 2019, 13:02:02 UTC |
a49c1a3 | Artem Artemev | 11 September 2019, 12:58:47 UTC | GPLVM models to GPflow2.0 | 11 September 2019, 12:58:47 UTC |
828694a | Sergio Diaz | 11 September 2019, 11:07:34 UTC | Tests for equivalence methods still in unsorted but updated to TF2 | 11 September 2019, 11:07:34 UTC |
5e0c087 | Sergio Diaz | 10 September 2019, 16:14:37 UTC | Merge branch 'awav/gpflow-2.0' into sergio_pasc/gpflow-2.0/move-coregion-tests | 10 September 2019, 16:14:37 UTC |
4ae9e39 | Artem Artemev | 10 September 2019, 16:12:04 UTC | GPMC, SGPMC and method.py test pass | 10 September 2019, 16:12:04 UTC |
b60b397 | Artem Artemev | 10 September 2019, 16:00:29 UTC | Merge branch 'sergio_pasc/gpflow-2.0/move-kldiv-tests' into awav/gpflow-2.0 | 10 September 2019, 16:00:29 UTC |
2dd2937 | Artem Artemev | 10 September 2019, 15:54:56 UTC | Merge branch 'sergio_pasc/gpflow-2.0/move-kldiv-tests' of github.com:GPflow/GPflow into sergio_pasc/gpflow-2.0/move-kldiv-tests | 10 September 2019, 15:54:56 UTC |
54357b0 | Artem Artemev | 10 September 2019, 15:54:34 UTC | Formatting | 10 September 2019, 15:54:34 UTC |
94e5329 | Sergio Diaz | 10 September 2019, 15:46:04 UTC | VGP coregion updated although test fail | 10 September 2019, 15:46:04 UTC |
01f5976 | Sergio Diaz | 10 September 2019, 15:35:28 UTC | Merge branch 'awav/gpflow-2.0' into sergio_pasc/gpflow-2.0/move-coregion-tests | 10 September 2019, 15:35:28 UTC |
6e1f56a | Artem Artemev | 10 September 2019, 15:23:06 UTC | Update VGP model | 10 September 2019, 15:23:06 UTC |
3b5c8c7 | Sergio Diaz | 10 September 2019, 13:57:09 UTC | Preparing Coregion tests for TF2.0 GPflow2.0 !!! Still don't pass (hence in _unsorted_tests folder), waiting for VGP to me merged | 10 September 2019, 13:57:09 UTC |
c5a1bc9 | Sergio Diaz | 10 September 2019, 13:27:38 UTC | Merge branch 'awav/gpflow-2.0' into sergio_pasc/gpflow-2.0/move-kldiv-tests # Conflicts: # _unsorted/_test_kldiv.py | 10 September 2019, 13:27:38 UTC |
bb09839 | Sergio Pascual | 10 September 2019, 09:26:31 UTC | SGPR and SGPFITC now have `data` as part of the state (#1043) | 10 September 2019, 09:26:31 UTC |
8394f5e | Artem Artemev | 09 September 2019, 23:31:12 UTC | Message in test_likelihood | 09 September 2019, 23:31:12 UTC |
cb7d952 | Artem Artemev | 09 September 2019, 22:52:19 UTC | Likelihood test pass at CircleCI | 09 September 2019, 22:52:19 UTC |
f666063 | Artem Artemev | 09 September 2019, 22:42:10 UTC | Debugging likelihood test | 09 September 2019, 22:42:10 UTC |
4e8906f | Artem Artemev | 09 September 2019, 21:37:32 UTC | Debugging likelihood tests | 09 September 2019, 21:37:32 UTC |
2d74480 | Artem Artemev | 09 September 2019, 21:31:50 UTC | Randomize order of tests | 09 September 2019, 21:31:50 UTC |
4864832 | Artem Artemev | 09 September 2019, 17:11:17 UTC | Pip update | 09 September 2019, 17:11:17 UTC |
212658e | Artem Artemev | 09 September 2019, 15:37:16 UTC | Trigger circleci build | 09 September 2019, 15:37:16 UTC |