提交 · f14dc05d6b77472cd0e69b4934e537c2ebbdd4da · testgroup / pytensor

15 9月, 2009 3 次提交
- removed perform from nll grad to work better with debugmode · f14dc05d
  由 James Bergstra 提交于 9月 14, 2009
  
  f14dc05d
- many fixes to GpuDownsample ops · 23535836
  由 James Bergstra 提交于 9月 14, 2009
  
  23535836
- new impl of GetSubtensor uses cuda_ndarray mapping protocol · ed85af6c
  由 James Bergstra 提交于 9月 14, 2009
  
  ed85af6c
29 8月, 2009 2 次提交
- added maxpooling op GpuDownsampleMax · cee77d13
  由 James Bergstra 提交于 8月 28, 2009
  
  cee77d13
- removed debugging assert · 3272fee0
  由 James Bergstra 提交于 8月 28, 2009
  
  3272fee0
28 8月, 2009 12 次提交
- removed whitespace · a6168f49
  由 James Bergstra 提交于 8月 27, 2009
  
  a6168f49
- merge · e380aa21
  由 James Bergstra 提交于 8月 27, 2009
  
  e380aa21
- added __str__ to nnet ops · 9d099f8f
  由 James Bergstra 提交于 8月 27, 2009
  
  9d099f8f
- added some comments · 10988f4f
  由 James Bergstra 提交于 8月 27, 2009
  
  10988f4f
- merge · c1765257
  由 James Bergstra 提交于 8月 27, 2009
  
  c1765257
- fix last commit and print less stuff. · 0e20b354
  由 Frederic Bastien 提交于 8月 27, 2009
  
  0e20b354
- lower the batch size of test_lenet_108 and test_lenet256 as the test was too long. · b6fa79b3
  由 Frederic Bastien 提交于 8月 27, 2009
  
  b6fa79b3
- added gpu implementation of Crossentropy gradient op · eb84fcb3
  由 James Bergstra 提交于 8月 27, 2009
  
  eb84fcb3
- added possibility to ignore_errors in run_test_conv_nnet2_classif to benchmark… · 85908603
  由 James Bergstra 提交于 8月 27, 2009
```
added possibility to ignore_errors in run_test_conv_nnet2_classif to benchmark even broken implementations.
```
  85908603
- Changed test_nnet to be able run the simple nnet in clinker mode. This revealed · dc678d1f
  由 James Bergstra 提交于 8月 27, 2009
```
that the opwiseclinker is efficient even for relatively small networks.
```
  dc678d1f
- fix test fct test_nnet.py:run_conv_nnet2_classif to don't have only 0 as target. · c750c40d
  由 Frederic Bastien 提交于 8月 27, 2009
  
  c750c40d
- modified elemwise to reuse output storage when possible. this sped it up · d7ca497f
  由 James Bergstra 提交于 8月 27, 2009
  
  d7ca497f
27 8月, 2009 3 次提交
- tweaked NaiveAlgo to be picklable · 7c8d604f
  由 James Bergstra 提交于 8月 26, 2009
  
  7c8d604f
- cleaned GpuElemwise __str__ since we are not using inplace pattern · 38c79505
  由 James Bergstra 提交于 8月 26, 2009
  
  38c79505
- added verbose flag to NaiveAlgo to control printing to stdout, stderr · 41dad32d
  由 James Bergstra 提交于 8月 26, 2009
  
  41dad32d
26 8月, 2009 3 次提交
- not counting scalars towards nd_collapse · 73e46362
  由 James Bergstra 提交于 8月 25, 2009
  
  73e46362
- converted to switch structure on nd_collapse · ac703c5b
  由 James Bergstra 提交于 8月 25, 2009
  
  ac703c5b
- prog on elemwise, but still very messy · 7a6b5233
  由 James Bergstra 提交于 8月 25, 2009
  
  7a6b5233
22 8月, 2009 5 次提交
- merge · bf825258
  由 James Bergstra 提交于 8月 21, 2009
  
  bf825258
- adding some new nnet sizes to optimize for · 954600b4
  由 James Bergstra 提交于 8月 21, 2009
  
  954600b4
- adding some faster Elemwise kernels, but more work to do still · 5aa6945a
  由 James Bergstra 提交于 8月 21, 2009
  
  5aa6945a
- print less stuff. · ab345880
  由 Frederic Bastien 提交于 8月 21, 2009
  
  ab345880
- reuse the size in the var instead of the hardcoded value. · e0b9af39
  由 Frederic Bastien 提交于 8月 21, 2009
  
  e0b9af39
21 8月, 2009 3 次提交
- fix typo. · fb47b71e
  由 Frederic Bastien 提交于 8月 21, 2009
  
  fb47b71e
- backport import to older version of python. · a75c39cc
  由 Frederic Bastien 提交于 8月 20, 2009
  
  a75c39cc
- allow to override the numpy include directory · 14fccb71
  由 Frederic Bastien 提交于 8月 20, 2009
  
  14fccb71
19 8月, 2009 9 次提交
- merge · a3758920
  由 James Bergstra 提交于 8月 18, 2009
  
  a3758920
- added a more naive elemwise implementation that seems just as fast as the… · bc8c8213
  由 James Bergstra 提交于 8月 18, 2009
```
added a more naive elemwise implementation that seems just as fast as the recusive-call implementation
```
  bc8c8213
- use the new ConvOp.set_flags() fct. · 50ffde6f
  由 Frederic Bastien 提交于 8月 18, 2009
  
  50ffde6f
- use the common parameter to don't duplicate them. · b5781703
  由 Frederic Bastien 提交于 8月 18, 2009
  
  b5781703
- bigger batch in test_nnet · cc0608b1
  由 James Bergstra 提交于 8月 18, 2009
  
  cc0608b1
- use the new hack in profile mode to print the MFlops of the op. · 6030673d
  由 Frederic Bastien 提交于 8月 18, 2009
  
  6030673d
- don't hide bug in profile mode. · 60b6facf
  由 Frederic Bastien 提交于 8月 18, 2009
  
  60b6facf
- cache compiled version of gpu code for faster compilation. · 1e8c3a25
  由 Frederic Bastien 提交于 8月 18, 2009
  
  1e8c3a25
- cache the gpu code for some op. · 61bf8215
  由 Frederic Bastien 提交于 8月 18, 2009
  
  61bf8215