Commit Graph
278 Commits
Author SHA1 Message Date
lameeus 64ee9acfdb 1) Extended layers to everything that has a ReLU activation
2) Added layer name to the warning
3) Corrected the dead neuron count (should work for both Theano and TF (tested for TF))
4) Added a print alongside the warning since for me the warning stopped showing after some epochs (might be a bug at my side).
2017-10-25 12:13:09 +02:00
Somshubra MajumdarandGitHub 43c21d9d8d Merge pull request #171 from seeb0h/master
Thanks for the PR!
2017-10-20 09:31:59 -05:00
Sébastieb Bosch 5af0db7178 Fix bad argument name for require_flatten 2017-10-20 15:59:03 +02:00
Andrew HundtandGitHub fd73547011 Merge pull request #169 from ahundt/remove_deconv3d
remove Deconvoluion3D, add missing backend functions
2017-10-18 02:10:47 -04:00
Andrew Hundt ad120f95e9 tensorflow_backend.py import dtype 2017-10-18 00:49:51 -04:00
Andrew Hundt 80d698c159 tensorflow_backend.py temporarily move _preprocess* functions into keras-contrib 2017-10-18 00:34:23 -04:00
Andrew Hundt 0d226a6775 remove more deconv3d code 2017-10-17 16:51:53 -04:00
Andrew Hundt a8ec7ad10a remove Deconvoluion3D because it is available in keras upstream
https://github.com/fchollet/keras/pull/7161
2017-10-17 16:45:55 -04:00
Andrew HundtandSomshubra Majumdar b0a2fbabfd resnet.py initial integration (#148)
Sorry for the delay.  

Looks much better now.
2017-10-15 20:21:50 -05:00
Andrew HundtandGitHub 7dc329a5a3 Merge pull request #162 from farizrahman4u/save_load_utils_tf_only
save_load_utils_test.py becomes tensorflow only to fix test failure until #157 is resolved
2017-10-15 14:23:42 -04:00
Andrew HundtandGitHub 70e68c11f5 save_load_utils_test.py make tensorflow only
Making this test tensorflow only until https://github.com/farizrahman4u/keras-contrib/issues/157 is resolved.
2017-10-15 13:28:43 -04:00
Andrew HundtandGitHub 99fe5e3b3a Merge pull request #142 from stygstra/batchrenorm-bugfix
Fix BatchRenormalization clipping (fixes #127)
2017-10-15 13:13:29 -04:00
Andrew HundtandGitHub 38937fdf7c Merge pull request #155 from farizrahman4u/setup
setup.py update to match keras
2017-10-15 13:04:53 -04:00
Andrew HundtandGitHub de768b6048 Merge pull request #161 from farizrahman4u/rm_test_file
save_load_utils_test.py removes 'model.h5'
2017-10-13 09:57:28 -04:00
David Stygstra 7577b8780a Fix BatchRenormalization clipping (fixes #127)
Previouly, a constant value was used for the clipping parameters. This
meant that it stayed at r=1 and d=0 forever, making it essentially
equivalent to regular batch normalization.
2017-10-11 18:45:35 -07:00
Andrew HundtandGitHub 6ec66fec74 save_load_utils_test.py removes 'model.h5' 2017-10-10 13:17:06 -04:00
Andrew HundtandGitHub a92268ffcb Merge pull request #146 from vilmar-hillow/master
Fix for Python 3 in conll dataset loading
2017-09-27 16:21:43 -04:00
Andrew HundtandGitHub 2e6b036052 Merge pull request #156 from farizrahman4u/recurrent
recurrent.py _time_distributed_dense doesn't exist
2017-09-25 14:44:36 -04:00
Andrew HundtandGitHub b2a05d7d1d recurrent.py pep8 2017-09-25 13:46:39 -04:00
Andrew HundtandGitHub 64734fbe03 recurrent.py _time_distributed_dense doesn't exist
upstream keras doesn't have the _time_distributed_dense symbol anymore
2017-09-25 13:40:47 -04:00
Andrew HundtandGitHub 48c840bda0 setup.py update to match keras 2017-09-25 13:27:02 -04:00
Somshubra MajumdarandGitHub 8d02cf2722 Merge pull request #151 from titu1994/fix-batchnorm
Fix normalization tests only. Remaining test still fail.
2017-09-22 15:41:22 -05:00
Somshubra Majumdar 821eff3948 fix normalization tests 2017-09-22 13:14:15 -05:00
Somshubra MajumdarandGitHub 97c31364cc Merge pull request #144 from titu1994/master
Refactor DenseNets
2017-09-19 08:40:18 -05:00
Somshubra Majumdar ff47da56fb Merge branch 'master' of https://github.com/titu1994/keras-contrib 2017-09-08 11:28:05 -05:00
Somshubra Majumdar 99c0c4b2d2 Remove dropout weight check 2017-09-08 11:27:56 -05:00
Somshubra MajumdarandGitHub 79c1bf7f26 Merge pull request #1 from ahundt/densenet_name_scope
Densenet name scope
2017-09-08 11:26:49 -05:00
Andrew Hundt b4dea23e65 cifar10_densenet.py fix parameter bug 2017-09-08 10:28:45 -04:00
Andrew Hundt b185d2655b densenet.py add name scope 2017-09-08 10:18:34 -04:00
Somshubra Majumdar 72f6ca1716 Update MD5 for weights 2017-09-07 08:50:05 -05:00
vilmar-hillow 14f3f3659b Fix for Python 3 in conll dataset loading 2017-09-07 15:19:48 +03:00
Somshubra Majumdar 5c6fd53a33 Fixed an incorrect passing of the dataformat to _preproce_input 2017-09-06 21:39:24 -05:00
Somshubra Majumdar 5bef6b9e52 Apply corrections 2017-09-06 21:33:24 -05:00
Somshubra Majumdar 7bae1db1f2 Update DenseNet with ImageNet weights and major code cleanup 2017-09-06 13:01:50 -05:00
Somshubra Majumdar 156c7e8e24 Correct dense connectivity in DenseNet 2017-09-01 18:14:24 -05:00
Michael OliverandGitHub c4d1a61746 fix missing paren 2017-08-16 21:01:26 -07:00
Michael OliverandGitHub 4ae3058c92 change np.exp to K.exp (#128)
* change np.exp to K.exp

To address https://github.com/farizrahman4u/keras-contrib/issues/127

* Update normalization.py

* Updates as suggested
2017-08-16 20:56:03 -07:00
Michael OliverandGitHub 5aed16cadc Update ftml for Keras changes to optimizers (#137)
* Update ftml for Keras changes to optimizers

* fix import
2017-08-16 20:51:32 -07:00
vfdevandMichael Oliver 157aa46f5a Fix typo 'tf' -> 'channels_last' (#135) 2017-08-16 17:48:37 -07:00
Michael Oliver 15133077af Fix test to refer to correct field for momentum weights in optimizer 2017-07-03 13:28:18 -07:00
Gleb SidoraandMichael Oliver 13ec4b4174 Added Dead ReLU detector callback (#115) 2017-06-30 15:08:39 -10:00
Antti TarvainenandMichael Oliver 8ec9c77382 Fix Pascal VOC installation for Python 3 (#114) 2017-06-21 09:42:47 -07:00
Andrew HundtandMichael Oliver ee6c396475 pascal_voc.py fix runtime errors (#112)
* pascal_voc.py fix runtime errors

* Fix PEP8
2017-06-19 15:25:57 -07:00
Mathieu Marquis BolducandMichael Oliver f0bb5becbb Instance Normalization layer (#101)
* Instance Normalization layer

* fix Instance Normalization for theano broadcasting rules

* Support rank 1 tensor in Instance Normalization

* address issue with axis parameter in instance normalization and add unit test for per channel correctness

* avoid assert_raises in normalization unit test

* Instance normalization set axis default to None
2017-06-19 08:57:54 -07:00
Michael Oliver 0b7eafa922 fix PEP8 2017-06-16 15:04:31 -07:00
Michael Oliver 6a8739b6c7 add ftml_test.py 2017-06-16 13:27:47 -07:00
Michael Oliver ef6c770639 add tests for ftml
add missing self.epsilon to ftml
2017-06-16 13:20:09 -07:00
Somshubra MajumdarandMichael Oliver b5669f4100 Fix training difficulty for DenseNetFCN (#96) 2017-06-16 11:29:20 -07:00
Shuai ZhengandMichael Oliver ebd568283a FTML optimizer (#110)
* add ftml.py and fix __init__.py

* fix PEP8
2017-06-16 11:28:43 -07:00
Michael Oliver a466050b5b add clip alias to correct place 2017-06-15 17:36:12 -07:00