提交 · d1f762aa9a2c74665a646bcb0da94d30802ca3ee · testgroup / pytensor

13 8月, 2014 34 次提交
- Add support for dimensions of size 1 in all cases. · d1f762aa
  由 Arnaud Bergeron 提交于 7月 30, 2014
  
  d1f762aa
- Remove the python version of these ops as it laughably slow and forces · 8e23c533
  由 Arnaud Bergeron 提交于 7月 30, 2014
```
a dependecy on scikits.cuda and pycuda.
```
  8e23c533
- Update docs to reflect batches and add some fallback code to add batches of 1 to… · d3088260
  由 Arnaud Bergeron 提交于 7月 22, 2014
```
Update docs to reflect batches and add some fallback code to add batches of 1 to non-batched version.
```
  d3088260
- Add batch support to blocksparse. · 42f4cb3e
  由 Arnaud Bergeron 提交于 7月 22, 2014
  
  42f4cb3e
- Use the right spelling for config.unittests.rseed. · 47d59687
  由 Arnaud Bergeron 提交于 7月 22, 2014
  
  47d59687
- Now the opt actually compute the right value and there is a test. · a7329037
  由 Arnaud Bergeron 提交于 7月 21, 2014
  
  a7329037
- Add optimizations to make the gradient update inplace. There are no tests yet. · f1515639
  由 Arnaud Bergeron 提交于 7月 21, 2014
  
  f1515639
- Add infer_shape to the ops. · 2e51a436
  由 Arnaud Bergeron 提交于 7月 21, 2014
  
  2e51a436
- Add C code using gemmBatched to SparseBlockDotOuterSS (the gradient). · 57865538
  由 Arnaud Bergeron 提交于 7月 17, 2014
  
  57865538
- Small forgotten speedup in SparseBlockDotGemvSS. · 98a15fa1
  由 Arnaud Bergeron 提交于 7月 17, 2014
  
  98a15fa1
- Use gemm_batched from python code in the gradient.ù · 9841c0db
  由 Arnaud Bergeron 提交于 7月 17, 2014
  
  9841c0db
- C code version of the python loop. · 437b1a5f
  由 Arnaud Bergeron 提交于 7月 17, 2014
  
  437b1a5f
- Remove leftover opt. · 69895eea
  由 Arnaud Bergeron 提交于 7月 16, 2014
  
  69895eea
- C code that uses SgemmBatched and a kernel to initialize the list of stuff. · ed244b6b
  由 Arnaud Bergeron 提交于 7月 16, 2014
  
  ed244b6b
- Fix memory leak in C code for blocksparse. · c774e32e
  由 Arnaud Bergeron 提交于 7月 14, 2014
  
  c774e32e
- Use gemm_batched in the python code. · 8f9c2a12
  由 Arnaud Bergeron 提交于 7月 14, 2014
  
  8f9c2a12
- Fix errors in C code and add a cache version. It passes the tests and works. · 29db8ffb
  由 Arnaud Bergeron 提交于 7月 14, 2014
  
  29db8ffb
- Add C code to SparseBlockGemvSS · 1519d758
  由 Arnaud Bergeron 提交于 7月 14, 2014
  
  1519d758
- Add support for the fortran order in gemv (and a test for it). · 0bc12fe9
  由 Arnaud Bergeron 提交于 7月 07, 2014
  
  0bc12fe9
- Fix shape error in tests (which also means that we had wrong behavior). · d12f4aea
  由 Arnaud Bergeron 提交于 7月 02, 2014
  
  d12f4aea
- And make the opt test work. · 7f15e04a
  由 Arnaud Bergeron 提交于 6月 25, 2014
  
  7f15e04a
- Add test to check that the inplace opts are working. · 6c77f4a6
  由 Arnaud Bergeron 提交于 6月 25, 2014
  
  6c77f4a6
- Don't try to use opt when cuda is not available. · 93879ae4
  由 Arnaud Bergeron 提交于 6月 25, 2014
  
  93879ae4
- Add inplace optimizations. · 1c3afdf6
  由 Arnaud Bergeron 提交于 6月 23, 2014
  
  1c3afdf6
- Use the non-scan version by default, since it's faster. · 6b162a92
  由 Arnaud Bergeron 提交于 6月 23, 2014
  
  6b162a92
- Fix last bug. · 9432f511
  由 Arnaud Bergeron 提交于 6月 23, 2014
  
  9432f511
- Fix the crash on exit with pycuda. · 4c101c36
  由 Arnaud Bergeron 提交于 6月 23, 2014
  
  4c101c36
- Fix the the gradient on W. Finally. · 6d327bd5
  由 Arnaud Bergeron 提交于 6月 23, 2014
  
  6d327bd5
- Add tests for grad shape. · 94fb4d03
  由 Arnaud Bergeron 提交于 6月 23, 2014
```
Some improvements to the gradient, but it's still transposed and partially wrong.
```
  94fb4d03
- Fix shape errors in the gradient (but it's still not the right value). · bcb902e9
  由 Arnaud Bergeron 提交于 6月 23, 2014
  
  bcb902e9
- Add tests for the grad of the op version (which is broken for now). · 09658800
  由 Arnaud Bergeron 提交于 6月 23, 2014
  
  09658800
- Add a test for the op version (but not for grad yet). · 493e71a4
  由 Arnaud Bergeron 提交于 6月 23, 2014
  
  493e71a4
- Address comments. · fabf1fdf
  由 Arnaud Bergeron 提交于 6月 20, 2014
  
  fabf1fdf
- First step, having something that works. · 4703a2b4
  由 Arnaud Bergeron 提交于 6月 19, 2014
  
  4703a2b4
12 8月, 2014 4 次提交
- Merge pull request #2031 from serge-m/master · 41be3b8b
  由 abergeron 提交于 8月 12, 2014
```
Fixed "cublasHandle_t handle" declaration in cuda_ndarray.cuh.
```
  41be3b8b
- Fixed "cublasHandle_t handle" declaration in cuda_ndarray.cuh. · 9973ad61
  由 Sergey Matyunin 提交于 8月 12, 2014
```
Required for cuda_convnet compilation in pylearn2 in Windows
```
  9973ad61
- Merge pull request #2021 from abergeron/pickle_as_op · 6c288a2a
  由 Pascal Lamblin 提交于 8月 11, 2014
```
Make as_op instances pickleable.
```
  6c288a2a
- Add a test for the pickle functionality. · d1e75733
  由 Arnaud Bergeron 提交于 8月 11, 2014
  
  d1e75733
08 8月, 2014 1 次提交
- Merge pull request #2016 from Tanjay94/SlowAttr2 · 674d54f5
  由 Frédéric Bastien 提交于 8月 08, 2014
```
SlowAttr2
```
  674d54f5
07 8月, 2014 1 次提交
- Merge pull request #2025 from shaibagon/patch-1 · 77a2f3aa
  由 Frédéric Bastien 提交于 8月 07, 2014
```
Improving documentation of max_pool_2d
```
  77a2f3aa