提交 · 740295e2f07561569b98702804a2eaa2f3c9841f · testgroup / pytensor

17 9月, 2014 2 次提交
- Make the core code for the convolution reusable. · 740295e2
  由 Arnaud Bergeron 提交于 9月 16, 2014
  
  740295e2
- Merge pull request #1967 from nouiz/fast_compile_gpu · c022347b
  由 abergeron 提交于 9月 16, 2014
```
[WIP] Fast compile gpu
```
  c022347b
16 9月, 2014 4 次提交
- Merge pull request #2103 from f0k/convop-grad-shapes · 1563ea38
  由 Frédéric Bastien 提交于 9月 16, 2014
```
Propagate shape information to ConvOp gradients
```
  1563ea38
- Merge pull request #2112 from nouiz/abergeron-dnn · 50968f7a
  由 Frédéric Bastien 提交于 9月 15, 2014
```
Abergeron dnn
```
  50968f7a
- Fix error handling. · 5b796440
  由 Frederic 提交于 9月 15, 2014
  
  5b796440
- Merge pull request #2100 from nouiz/omlw · fa1d9c4a
  由 Frédéric Bastien 提交于 9月 15, 2014
```
Open Machine Learning Workshop 2014 presentation
```
  fa1d9c4a
15 9月, 2014 2 次提交
- add check before patternbroadcast in local_gpualloc_memset_0 · 1acdafac
  由 f0k 提交于 9月 15, 2014
  
  1acdafac
- pep8 · 08c6d742
  由 f0k 提交于 9月 15, 2014
  
  08c6d742
13 9月, 2014 1 次提交
- Merge pull request #1936 from daemonmaker/issue1569 · d12a03be
  由 Frédéric Bastien 提交于 9月 12, 2014
```
Added to test to determine the behavior of the BLAS and intialize the me...
```
  d12a03be
12 9月, 2014 25 次提交
- Merge pull request #2107 from vdumoulin/docfix · 0f1c5ac3
  由 Frédéric Bastien 提交于 9月 12, 2014
```
keep_dims -> keepdims in theano.tensor.nnet documentation
```
  0f1c5ac3
- keep_dims -> keepdims in theano.tensor.nnet documentation · 90397682
  由 Vincent Dumoulin 提交于 9月 12, 2014
  
  90397682
- gpuarray opt to move shape to the GPU. · c4fed2b2
  由 Frederic 提交于 9月 11, 2014
```
I'm not able to reproduce the problem. It happen in the DLT
logistic_sgd with fast_compile optimizer.
```
  c4fed2b2
- make fast_compile work with the new gpu back-end. · e5e40aec
  由 Frederic 提交于 9月 10, 2014
  
  e5e40aec
- Fix comment about code habit · c10087da
  由 Frederic 提交于 9月 10, 2014
  
  c10087da
- Fixed return statement. · 4c638396
  由 Dustin Webb 提交于 9月 09, 2014
  
  4c638396
- Reworked register_stabilize to work like register_canonicalize and register_specialize. · 1d8302e5
  由 Dustin Webb 提交于 9月 09, 2014
  
  1d8302e5
- Reworked opt.register_specialize to match the way opt.register_canonicalize works. · f018f34a
  由 Dustin Webb 提交于 9月 09, 2014
  
  f018f34a
- Changed name of parameter of inner function to disambiguate it. · f34caad6
  由 Dustin Webb 提交于 9月 03, 2014
  
  f34caad6
- Combined register_canonicalize_tags and register_canonicalize. · ab3bce3f
  由 Dustin Webb 提交于 8月 29, 2014
  
  ab3bce3f
- Added a new decorator and tagged a few opts for optimization when using the gpu… · b94282cb
  由 Dustin Webb 提交于 8月 18, 2014
```
Added a new decorator and tagged a few opts for optimization when using the gpu or optimizer=fast_compile.
```
  b94282cb
- fix import crash. · 67c0dfea
  由 Frederic 提交于 7月 21, 2014
  
  67c0dfea
- Register to the gpu tag more opt, as we don't have all the other case on the GPU… · 705806f4
  由 Frederic 提交于 7月 21, 2014
```
Register to the gpu tag more opt, as we don't have all the other case on the GPU otherwise in fast_compile.
```
  705806f4
- Fix crash in the optimizer profiler. · 9de8a09d
  由 Frederic 提交于 4月 10, 2014
  
  9de8a09d
- Fix the opt local_gpu_dot_to_dot22 to move more dot to the GPU in · c930d96a
  由 Frederic 提交于 4月 09, 2014
```
fast_compile.
```
  c930d96a
- Make sure dot(matrix, matrix) get moved to the gpu in fast_compile · f8a19f5b
  由 Frederic 提交于 4月 09, 2014
  
  f8a19f5b
- Also remove useless op in fast_run. · 8c7991b3
  由 Frederic 提交于 4月 09, 2014
  
  8c7991b3
- Don't make gpuelemwise inplace opt enabled with device=gpu and optimizer=fast_compile. · b19eee60
  由 Frederic 提交于 4月 09, 2014
  
  b19eee60
- Add fast_compile to GPU opt that move computation to the GPU. · 4897207a
  由 Frederic 提交于 4月 09, 2014
  
  4897207a
- Merge pull request #2105 from carriepl/profilingOutput · aa6066ca
  由 abergeron 提交于 9月 11, 2014
```
 Add new flag for theano profiling output cropping
```
  aa6066ca
- Add documenation on how to use GpuDnnConv. · 74bc7ab6
  由 Arnaud Bergeron 提交于 9月 11, 2014
  
  74bc7ab6
- Check that the inputs are contiguous. · 4e634669
  由 Arnaud Bergeron 提交于 9月 11, 2014
  
  4e634669
- Add the cudnn error message to the error strings. · 1a593ff3
  由 Arnaud Bergeron 提交于 9月 11, 2014
  
  1a593ff3
- Fix bad fail code for variable declarations in the struct. · a4c65bf1
  由 Arnaud Bergeron 提交于 9月 11, 2014
  
  a4c65bf1
- Added flag profiling.output_line_width to limit line max length of profiling output · f40852da
  由 Pierre Luc Carrier 提交于 9月 10, 2014
  
  f40852da
11 9月, 2014 6 次提交
- Fix ConvOp shape inference and propagation for imshp=None · c78d5c22
  由 f0k 提交于 9月 11, 2014
  
  c78d5c22
- make local_gpu_alloc_memset_0() keep the original broadcast pattern · 20e7ec9d
  由 f0k 提交于 9月 11, 2014
  
  20e7ec9d
- Simplify error handling in struct init code and stop ignoring errors. · e2d4b31f
  由 Arnaud Bergeron 提交于 9月 10, 2014
  
  e2d4b31f
- Corrected comment. · 7790b329
  由 Dustin Webb 提交于 9月 10, 2014
  
  7790b329
- Make sure to initialize the cudnn descriptors. · 95afba78
  由 Arnaud Bergeron 提交于 9月 10, 2014
  
  95afba78
- Make ConvOp propagate any available shape information to instantiated gradient ops · 0c3ab9a5
  由 f0k 提交于 9月 10, 2014
  
  0c3ab9a5