提交 · 3007bf7949d7cd67af0cc9402e344d7fea9effc6 · testgroup / pytensor

18 10月, 2016 2 次提交
- Merge pull request #4915 from abergeron/dnn_rnn2 · 3007bf79
  由 Pascal Lamblin 提交于 10月 17, 2016
```
Cudnn RNN bindings.
```
  3007bf79
- Merge pull request #5103 from nouiz/floatX_error · d72325e4
  由 Pascal Lamblin 提交于 10月 17, 2016
```
Make device=gpu with floatX=float16 raise an error (cause many user q…
```
  d72325e4
17 10月, 2016 2 次提交
- Make device=gpu with floatX=float16 raise an error (cause many user questions… · 301f7a2f
  由 Frederic Bastien 提交于 10月 17, 2016
```
Make device=gpu with floatX=float16 raise an error (cause many user questions and this should never be used). if floatX=float64, raise an error.
```
  301f7a2f
- Merge pull request #5096 from nouiz/fast_compile · 173826a7
  由 Frédéric Bastien 提交于 10月 17, 2016
```
fix tests in mode=FAST_COMPILE
```
  173826a7
16 10月, 2016 1 次提交
- Merge pull request #5069 from gvtulder/f-gpuarray-corr_gemm · 0acf8988
  由 Pascal Lamblin 提交于 10月 15, 2016
```
GpuCorrMM and GpuCorr3dMM in new backend
```
  0acf8988
15 10月, 2016 35 次提交
- Fix flake8 error that slipped in the rebase. · c27959ba
  由 Arnaud Bergeron 提交于 10月 14, 2016
  
  c27959ba
- Fix the grad_c test. · 8cd94e47
  由 Arnaud Bergeron 提交于 10月 14, 2016
  
  8cd94e47
- Fix the connectivty test. · 3d85d7cf
  由 Arnaud Bergeron 提交于 10月 14, 2016
  
  3d85d7cf
- Flake8 · b0982c25
  由 Arnaud Bergeron 提交于 10月 14, 2016
  
  b0982c25
- Fix test. · 242145f6
  由 Arnaud Bergeron 提交于 10月 14, 2016
  
  242145f6
- Add tests for gradient connectivity to cy only. · f722afba
  由 Arnaud Bergeron 提交于 10月 14, 2016
  
  f722afba
- Add tests for various patterns of connectivity of y and hy. · fd2f23f3
  由 Arnaud Bergeron 提交于 10月 14, 2016
  
  fd2f23f3
- Remove the dropout arguemnts since they aren't supported. · 8062f2a9
  由 Arnaud Bergeron 提交于 10月 14, 2016
  
  8062f2a9
- Replace dy with zeros if disconnected instead of dhy[-1]. · 9a0208da
  由 Arnaud Bergeron 提交于 10月 14, 2016
  
  9a0208da
- After a re-read of the docs, adjust the shape comments. · dd9de351
  由 Arnaud Bergeron 提交于 10月 14, 2016
  
  dd9de351
- Flake8 fix. · 5fc707eb
  由 Arnaud Bergeron 提交于 9月 29, 2016
  
  5fc707eb
- Fix LSTM test. · 612ea55d
  由 Arnaud Bergeron 提交于 9月 29, 2016
  
  612ea55d
- Fix segfault. · ce0494c9
  由 Arnaud Bergeron 提交于 9月 28, 2016
  
  ce0494c9
- Add a test for an LSTM variant. · 3e4b1d77
  由 Arnaud Bergeron 提交于 9月 28, 2016
  
  3e4b1d77
- Fix grad · 01305008
  由 Arnaud Bergeron 提交于 9月 28, 2016
  
  01305008
- Fix some things and remove extraneous dimshuffles. · e57898bc
  由 Arnaud Bergeron 提交于 9月 27, 2016
  
  e57898bc
- Rename grad2 to L_op. · 7b23b23b
  由 Arnaud Bergeron 提交于 9月 27, 2016
  
  7b23b23b
- Fix error message for cudnn library not found. · 5b10cc54
  由 Arnaud Bergeron 提交于 9月 26, 2016
  
  5b10cc54
- Some changes from review. · d6903d00
  由 Arnaud Bergeron 提交于 9月 21, 2016
  
  d6903d00
- Fix flags for split matrices. · ab154880
  由 Arnaud Bergeron 提交于 9月 20, 2016
  
  ab154880
- We need to set dw to zeros since it adds to it. · 2ec9998a
  由 Arnaud Bergeron 提交于 9月 19, 2016
  
  2ec9998a
- Add a test for gru mode of cudnn. · d7bf519a
  由 Arnaud Bergeron 提交于 9月 16, 2016
  
  d7bf519a
- Make CDataType work under compute_test_value. · 61554830
  由 Arnaud Bergeron 提交于 9月 16, 2016
  
  61554830
- Make change_flags also work as a context manager. · 0850be0d
  由 Arnaud Bergeron 提交于 9月 16, 2016
  
  0850be0d
- Fix the strides for the weight matrices. · 1f01eab6
  由 Arnaud Bergeron 提交于 9月 15, 2016
  
  1f01eab6
- v5rc has different prototypes for some versions so reject it. · 32b75e7f
  由 Arnaud Bergeron 提交于 9月 13, 2016
  
  32b75e7f
- Align with the commited names. · aab7cead
  由 Arnaud Bergeron 提交于 9月 13, 2016
  
  aab7cead
- Use __props__ for CDataType and add the missing items. · 94199f17
  由 Arnaud Bergeron 提交于 9月 06, 2016
  
  94199f17
- Try to give better shape information about params. · 192b0e47
  由 Arnaud Bergeron 提交于 9月 06, 2016
  
  192b0e47
- Add a way to split the params into components. · d1e64669
  由 Arnaud Bergeron 提交于 9月 02, 2016
```
This is still less than useful since we only get the size of each
component without any indications of the dimensions of matrices.
```
  d1e64669
- Fixes from review. · 17a0327b
  由 Arnaud Bergeron 提交于 8月 31, 2016
  
  17a0327b
- Add ops and support code for the RNN binding for cudnn. · 77cc5729
  由 Arnaud Bergeron 提交于 8月 30, 2016
  
  77cc5729
- Create the cudnn handle on context initialization and pass it around. · 4b94a811
  由 Arnaud Bergeron 提交于 8月 30, 2016
  
  4b94a811
- Add a way to set properties on a context. · 3f536fb7
  由 Arnaud Bergeron 提交于 8月 30, 2016
  
  3f536fb7
- Add grad2 method. · caa214c7
  由 Arnaud Bergeron 提交于 8月 30, 2016
  
  caa214c7