Commit graph

1824 commits

Author SHA1 Message Date
Jerry Wu
69fe0e2722 support layerwise serialization (#1472)
* support layerwise serialization

* support layerwise serialization

* add unit tests

* remove commented code

* fix style issue

* change invoke api

* add new api

* resolve conflict

* fix typo

* refinement per review

* refinement per review

* refinement
2017-08-18 17:06:11 +08:00
Yan Wan
2a29808ffc add getTimes in recurrent and TimeDistributed layer (#1470)
* add getTimes

* add getTimes for all Cells

* addTime method in cell

* add appendTimes method

* remove nullException when frequently calling setTimes and getTimes in recurrent layer

* fix bugs in calculating backwardSum

* add getTimes for Graph

* revise ConcatTable

* remove add in calculating backwardTime
2017-08-17 16:24:28 +08:00
zhangxiaoli73
377d90cd4d optimize Sigmoid (#1471) 2017-08-17 13:43:41 +08:00
Yang Wang
0f372cd19c support same padding in maxpooling and avgpooling (#1458)
* support same padding

* add doc

* refactor

* add doc
2017-08-17 10:06:11 +08:00
Yao Zhang
c358955cef Add abstract Operation, Add, Substract, Multiply and Divide operations (#1462)
* finsh Operation, Add, Substract, Multiply and Divide operations

* meet code review

* make acc not be supported

* meet code review

* make documentation better

* rename package object

* rename package object
2017-08-16 17:12:24 +08:00
zhangxiaoli73
058a1cac5e add mkl.tanh (#1450)
* add mkl.tanh

* change core

* chang tanh
2017-08-16 15:54:39 +08:00
Yan Wan
6baef96116 remove set operation (#1464) 2017-08-16 10:51:55 +08:00
Yang Wang
76007c1a30 support nhwc format in SpatialConvolution (#1434)
* support nhwc format in SpatialConvolution

* add python api and doc

* meet code review

* fix bug and add tests

* meet code review

* fix bug
2017-08-14 05:55:26 -07:00
Yan Wan
7d8612d806 optimize logsoftmax (#1424)
* optimize logsoftmax

* use ev instead of MKL

* add clearState

* revise precision of InceptionSpec and EvaluatorSpec using MKL

* revise GoogleNet_v1 test precision

* change buffer name
2017-08-14 15:25:24 +08:00
Yan Wan
78e5653d2a fix copyStatus in TimeDistributed layer (#1451) 2017-08-14 14:05:41 +08:00
zhangxiaoli73
7ac4bbfcc5 add funcional model definition (#1200)
* add inception and lenet graph

* add resnet,autoencoder,alexnet graph

* fix graph unit test

* add vgg graph and test

* add inception_v2 graph

* add inception-v1 graph

* add graph model example

* change JoinTable nInputDims
2017-08-14 11:05:38 +08:00
Yan Wan
59aa377c8b revise LSTMPeephole to Graph (#1430) 2017-08-11 15:07:57 +08:00
Yan Wan
914216b258 GRU to Graph (#1429)
* GRU to Graph

* set withBias = false in gru h2g
2017-08-11 13:57:28 +08:00
Yan Wan
ce22aae616 SimpleRNN to Graph (#1431)
* SimpleRNN to Graph

* move i2h, h2h to buildGraph in SimpleRNN
2017-08-11 13:57:10 +08:00
Yan Wan
86371f3907 LSTM to Graph (#1428) 2017-08-11 13:56:37 +08:00
dding3
38251491a6 Set hiddenstate (#1439)
* support set/get state in recurrent
2017-08-10 20:34:53 -07:00
Yang Wang
fcdb4f0355 add nhwc im2col and col2im kernel (#1419)
* add nhwc im2col and col2im kernel

* address comments
2017-08-07 16:58:46 +08:00
Yao Zhang
3e610768be Add 3D de-convolution (#1401)
* finish volumetric full convolution

* add documentation

* reset core

* meet code review

* add serial version uid

* meet code review
2017-08-03 11:04:21 +08:00
Xin Qiu
2aa51233a8 Check target size in ClassNll (#1420) 2017-08-02 14:27:18 +08:00
dding3
fc4fd2005b add regularizer for convlstmpeephole3d (#1395)
* add regularizer for convlstmpeephole3d
* rename convlstmpeephole2d back to convlstmpeephole
2017-07-27 23:10:48 -07:00
Xianyan
1713f68127 add regulaizer to volumetricconvolution (#1371)
* add regulaizer to volumetricconvolution

* fix name

* update doc

* update doc
2017-07-27 16:45:20 +08:00
Jerry Wu
4e5e25662a Fix temporalconv (#1359)
* add dynamic caffe loading support

* refined with bug fix and more commnets

* fix typo and bias check for convolution

* renaming change

* fix import reference

* refine customized layer support

* fix style check issue

* resolve style issue

* resolve conflict

* add missing method
2017-07-27 14:49:26 +08:00
dding3
4bf88dacb5 support get final state and cell (#1369)
* support get final state and cell
2017-07-26 22:37:11 -07:00
Xianyan
61d443e51c add SpatialWithinChannelLRN, update the mapping of caffe lrn (#1364)
* add SpatialWithinChannelLRN

* add python

* fix map error

* fix python

* meet code review

* update doc
2017-07-26 16:18:23 +08:00
Xianyan
9757dbd1a7 Support global average pooling (#1365)
* add global pooling

* update doc and python

* update doc
2017-07-26 15:35:22 +08:00
dding3
543a3762af Convlstm3d (#1361)
* support convlstm3d
2017-07-25 22:17:34 -07:00
dding3
13ea525adc remove unnecessary contiguous check in volumetricconvolution (#1357)
remove unnecessary contiguous check in volumetricconvolution
2017-07-24 20:59:26 -07:00
Yan Wan
0ed660313a Recurrent: fix clearState (#1306) 2017-07-20 14:43:13 +08:00
Ian Wong
ea3f8cf661 Add scala tensorflow load/save examples (#1291)
* add scala tf example

* update

* update

* fix a small typo

* fix unit test
2017-07-20 00:39:06 -05:00
Yan Wan
dc0d060505 fix batchSize non-consistency bug (#1295)
add unit test
2017-07-20 12:36:13 +08:00
Yan Wan
b51c466961 add CosineDistanceCriterion (#1195)
* add CosineDistanceCriterion

* add cosineDistanceCriterion

* add CosineDistanceCriterion python test

* add python test

* add blank line in pythonapi

* revise APIDocs
2017-07-19 20:28:29 +08:00
Yang Wang
7772dece03 Refine tensorflow doc (a follow up of pr #1253 to meet code review) (#1272)
* meet code review

* fix example

* meet code review

* refine title

* make saveTF compatible with saveCaffe
2017-07-18 22:45:28 -07:00
Ian Wong
e16bfe4599 Graph doc (#1275)
* Add Input docs

* fix bugs

* meet code review
2017-07-19 00:19:56 -05:00
Xin Qiu
c91f5242ab update shared convolution (#1144)
update shared convolution with latest SpatialConvolution.
Add wRegularize, bRegularizer, initWeight, initBias, initGradWeight, initGradBias, withBias
2017-07-19 08:39:30 +08:00
Xin Qiu
f658e378ce fix SpatialConvolution's hashcode and clearState will throw NullPointerException (#1268) 2017-07-18 15:24:54 +08:00
Xianyan
7aa7bd16e8 add load caffe model doc, remove useless parameter, add python api (#1266)
* add load caffe model doc, remove useless parameter, add python api

* update doc

* add save caffe

* remove empty lines

* fix python test

* Update caffe-support.md

* fix python

* update doc
2017-07-18 15:10:31 +08:00
zhangxiaoli73
9c449c1ac7 optimize joinTable perf (#1248) 2017-07-17 23:13:43 -05:00
Xianyan
1c2d630932 Fix the bug of JoinTable after calling clearState() (#1257) 2017-07-18 10:28:47 +08:00
Yang Wang
dfdb03cc17 fix concat when the dimension concatenated along is larger than 2 (#1230)
* fix concat when dim >= 3

* fix backward

* fix backward

* changing back to tensor copy

* meet code review
2017-07-13 23:29:36 -07:00
ding
ee994fb14d convert back convlstm optimization 2017-07-13 16:05:24 -07:00
Yan Wan
df873d648c BatchNormalization accepts default values (#1216)
* BatchNormalization accepts default values

* runningMean defined during run time

* remove multiple constructor in BatchNormalization class
2017-07-13 15:17:57 +08:00
Xin Qiu
5b665b64cf fix inception and xavier (#1220) 2017-07-13 12:18:09 +08:00
Yao Zhang
d733f90901 Tree LSTM with Sentiment classification example (#1217)
* finish treeLSTM with sentiment example

* fix scala style

* fix python error

* fix a python bug

* fix a typo

* fix a typo
2017-07-13 06:23:49 +08:00
Yao Zhang
1033e735b7 Make ClassNLL can process -1 label for padding, not update the model. (#1215)
* add -1 for ClassNLL

* fix typo

* add tests
2017-07-12 17:21:31 +08:00
jenniew
52fc2bcba5 update local jvm with comments 2017-07-12 02:02:22 -07:00
jenniew
1373b79587 update abstract module 2017-07-12 02:02:22 -07:00
jenniew
76411f929a create localModule and update local predictor 2017-07-12 02:02:22 -07:00
Yao Zhang
98717d6d8b Fix maptable bug (#1214)
* fix maptable bug

* Add new line to end of maptable
2017-07-12 16:44:41 +08:00
Yao Zhang
9909c1e630 Add 1D Convolution (TemporalConvolution) (#1180)
* add test

* remove unnecessary change in LinearSpec

* finish Temporal Convolution

* add python API

* meet code review

* meet code review

* fix typos

* fix typos

* meet code review
2017-07-12 15:51:00 +08:00
Yao Zhang
fe05012ea9 Fix a bug in the MV layer (#1171)
* add test

* remove unnecessary change in LinearSpec

* fix MV bugs
2017-07-12 14:28:31 +08:00