Matt Graham
|
f13ba91f82
|
Dealing with case where number of epochs less than stats interval.
|
2016-10-17 23:32:23 +01:00 |
|
Matt Graham
|
43ad2a00ba
|
Adding RBF layer definition for 4th lab.
|
2016-10-14 03:36:46 +01:00 |
|
Matt Graham
|
2e5a1a4c4f
|
Renaming MNIST eval set to test set for consistency with slides.
|
2016-10-14 03:36:16 +01:00 |
|
Matt Graham
|
4c4cf2cf27
|
Bug fix: incorrect use of zip in MomentumLearningRule.
|
2016-10-12 23:11:38 +01:00 |
|
Matt Graham
|
a6fcdc2894
|
Fixing typo in initialiser assert.
|
2016-10-10 10:09:29 +01:00 |
|
Matt Graham
|
cfad36c91f
|
Removing final backpropagation of gradients to inputs as unnecessary.
|
2016-10-10 09:27:02 +01:00 |
|
Matt Graham
|
a24ff90f71
|
Adding Nesterov momentum learning rule.
|
2016-10-10 09:25:52 +01:00 |
|
Matt Graham
|
259b000cba
|
Adding schedulers to optimiser.
|
2016-10-10 09:25:33 +01:00 |
|
Matt Graham
|
18f893d864
|
Adding Glorot and Bengio parameter initialiser.
|
2016-10-10 09:24:32 +01:00 |
|
Matt Graham
|
302d44371a
|
Adding new schedulers module.
|
2016-10-10 09:23:53 +01:00 |
|
Matt Graham
|
fa2f6b195c
|
Correcting docstring typo.
|
2016-10-08 22:19:49 +01:00 |
|
Matt Graham
|
25263e1391
|
Including initial pre-training statistics.
|
2016-10-07 06:23:15 +01:00 |
|
Matt Graham
|
e05717e3d0
|
Adding SoftmaxLayer definition.
|
2016-10-07 06:22:12 +01:00 |
|
Matt Graham
|
8d6f37668f
|
Adding ability to adjust batch size after initialisation and resetting to initial state (order).
|
2016-10-07 06:21:34 +01:00 |
|
Matt Graham
|
4ef1428447
|
Switching from 'cost' to 'error' for consistency with slides.
|
2016-09-30 02:53:13 +01:00 |
|
Matt Graham
|
2f9253bd21
|
Adding CCPP data and data provider.
|
2016-09-28 05:07:01 +01:00 |
|
Matt Graham
|
049ce20d71
|
Adjusting comment spacing.
|
2016-09-22 15:06:20 +01:00 |
|
Matt Graham
|
0bfe0c1a34
|
Documenting optimiser module.
|
2016-09-21 00:54:36 +01:00 |
|
Matt Graham
|
dac0729324
|
Adding docstrings to costs and removing accuracies.
|
2016-09-21 00:54:21 +01:00 |
|
Matt Graham
|
74ddb82dcb
|
Adding documentation to models module.
|
2016-09-20 23:09:27 +01:00 |
|
Matt Graham
|
9d3ee17418
|
Adding documentation to learning rules.
|
2016-09-20 14:48:56 +01:00 |
|
Matt Graham
|
48b02abf39
|
Adding more documentation to data providers module.
|
2016-09-20 12:49:42 +01:00 |
|
Matt Graham
|
f02e48dbac
|
Renaming trainers module to optimisers.
|
2016-09-19 12:25:43 +01:00 |
|
Matt Graham
|
2988811b50
|
Removing old optimisers code.
|
2016-09-19 12:22:47 +01:00 |
|
Matt Graham
|
03f27cab75
|
Added ability to add additional monitoring channels during training.
|
2016-09-19 12:18:44 +01:00 |
|
Matt Graham
|
d55311e501
|
Renaming data providers module for consistency.
|
2016-09-19 11:17:42 +01:00 |
|
Matt Graham
|
3ad26049ba
|
Cleaning up and refactoring data providers.
|
2016-09-19 11:16:53 +01:00 |
|
Matt Graham
|
08c5bcbce4
|
Tidying up headers.
|
2016-09-19 11:16:21 +01:00 |
|
Matt Graham
|
e9502c44a4
|
Moving default seed to package level and adding author metadata.
|
2016-09-19 11:14:10 +01:00 |
|
Matt Graham
|
8c98867cfa
|
Removing convolution placeholders.
|
2016-09-19 07:33:37 +01:00 |
|
Matt Graham
|
89beb62fce
|
Initial refactor of framework.
|
2016-09-19 07:31:31 +01:00 |
|
Pawel Swietojanski
|
d8c7ae96a0
|
fix to bias shape assert
|
2015-12-28 16:15:04 +00:00 |
|
Matt Graham
|
5e20946f20
|
Fixing incorrect numpy references in ConvLinear tests.
|
2015-12-22 07:39:09 +00:00 |
|
Matt Graham
|
eef54933c1
|
Adding ConvLinearLayer test functions.
|
2015-12-20 20:24:21 +00:00 |
|
pswietojanski
|
57b7496083
|
Merge remote branch 'upstream/master'
|
2015-12-13 20:08:12 +00:00 |
|
pswietojanski
|
ca8c429973
|
fix to dropout scaling
|
2015-12-13 19:53:10 +00:00 |
|
Matt Graham
|
efd6d49fa1
|
Making schedulers get_rate interface consistent with use in optimisers (i.e. gets passed accuracy not error rate).
|
2015-12-01 15:04:50 +00:00 |
|
pswietojanski
|
9969b47ba4
|
fixing naming of bprop_cost (thanks to anyone who reported this to Matt)
|
2015-11-17 13:59:16 +00:00 |
|
pswietojanski
|
c77bc3b1cd
|
PEP8 normalisation, better Maxout comments
|
2015-11-16 14:04:47 +00:00 |
|
pswietojanski
|
16704f6471
|
bugfix to dropout (wrong scaling)
|
2015-11-16 13:55:10 +00:00 |
|
pswietojanski
|
bfd84f3ff6
|
fix to wrong tabulation
|
2015-11-15 21:53:47 +00:00 |
|
pswietojanski
|
b291e0dbd1
|
solutions 04 and 05
|
2015-11-15 21:36:35 +00:00 |
|
pswietojanski
|
689b78e4f4
|
adding missing conv compatibility code in fc linear transform
|
2015-11-15 16:45:41 +00:00 |
|
pswietojanski
|
fc2eecf135
|
adding overlooked dataset changes
|
2015-11-15 16:41:21 +00:00 |
|
pswietojanski
|
ed47b36873
|
coursework2, labs and code
|
2015-11-15 16:33:53 +00:00 |
|
pswietojanski
|
cb6712578d
|
solutions to 04 and 05
|
2015-11-15 16:00:58 +00:00 |
|
pswietojanski
|
c130785b79
|
some progress with 05 solutions
|
2015-11-14 17:06:12 +00:00 |
|
Pawel Swietojanski
|
07f0fa8644
|
fix to L1 prior cost computation
|
2015-11-03 14:08:23 +00:00 |
|
pswietojanski
|
d87f8b05aa
|
some clarifications
|
2015-11-02 12:59:36 +00:00 |
|
pswietojanski
|
faaa6cb172
|
lab 4
|
2015-11-01 19:24:35 +00:00 |
|