提交 · 79c950faf03fcbb74f1be1804643bcb28477c664 · testgroup / pytensor

07 9月, 2012 28 次提交
- added a unit test that gradient with respect to int matches gradient · 79c950fa
  由 Ian Goodfellow 提交于 9月 05, 2012
```
with respect to float
```
  79c950fa
- added checks to grad for sigmoid · 7f85d302
  由 Ian Goodfellow 提交于 9月 05, 2012
  
  7f85d302
- corrected grad for dot · 7966b05c
  由 Ian Goodfellow 提交于 9月 05, 2012
  
  7966b05c
- corrected grad for permute · 049edc8d
  由 Ian Goodfellow 提交于 9月 05, 2012
  
  049edc8d
- implemented connection_pattern and corrected grad for Reshape · 94aa3737
  由 Ian Goodfellow 提交于 9月 05, 2012
  
  94aa3737
- implemented correction_pattern and corrected grad for Subtensor · 4990aaf9
  由 Ian Goodfellow 提交于 9月 05, 2012
  
  4990aaf9
- corrected SpecifyShape.grad · d1879e8b
  由 Ian Goodfellow 提交于 9月 05, 2012
  
  d1879e8b
- added "node" argument to ops that already had a connection pattern · e205110f
  由 Ian Goodfellow 提交于 9月 05, 2012
  
  e205110f
- gave connection_pattern a "node" argument to support ops with · f08fe14e
  由 Ian Goodfellow 提交于 9月 05, 2012
```
variable #s of inputs
```
  f08fe14e
- added test that the upgrade of connection_pattern works correctly · 71a8d3aa
  由 Ian Goodfellow 提交于 9月 05, 2012
  
  71a8d3aa
- fixed a test that relied on the grad method being allowed to return None · 20a1aee0
  由 Ian Goodfellow 提交于 9月 05, 2012
  
  20a1aee0
- upgraded shape op's connection_pattern to the new type · cdafffc7
  由 Ian Goodfellow 提交于 9月 05, 2012
```
made elemwise ops handle DisconnectedType correctly
```
  cdafffc7
- made gradient.grad use the upgraded connection_type method · bf0e90ff
  由 Ian Goodfellow 提交于 9月 05, 2012
  
  bf0e90ff
- documented upgrade of connection_pattern · ad865eb3
  由 Ian Goodfellow 提交于 9月 05, 2012
  
  ad865eb3
- added unit test to make sure connection pattern of shape and alloc work · 742b27bb
  由 Ian Goodfellow 提交于 9月 05, 2012
```
right
```
  742b27bb
- fixed grad method of shape op · 2c2ce0d2
  由 Ian Goodfellow 提交于 9月 05, 2012
  
  2c2ce0d2
- made true div return 0 gradient if its output is an integer and · e571b789
  由 Ian Goodfellow 提交于 9月 05, 2012
```
a floating point gradient otherwise
```
  e571b789
- fixed bug where tensor.grad could try to add DisconnectedType to other · 0e7eb0cc
  由 Ian Goodfellow 提交于 9月 05, 2012
```
variables
```
  0e7eb0cc
- fixed Alloc's grad method · dc6ecb37
  由 Ian Goodfellow 提交于 9月 05, 2012
  
  dc6ecb37
- fixed grad method for Add · 7b432b5a
  由 Ian Goodfellow 提交于 9月 06, 2012
  
  7b432b5a
- Merge pull request #909 from mrocklin/print-device-flag · ef8aa276
  由 nouiz 提交于 9月 06, 2012
```
add print_active_device flag
```
  ef8aa276
- add print_active_device flag · 008ebfaa
  由 Matthew Rocklin 提交于 9月 06, 2012
  
  008ebfaa
- Merge pull request #908 from lamblin/fix_test_neib · ca55d2f9
  由 nouiz 提交于 9月 06, 2012
```
Fix tests for image2neibs
```
  ca55d2f9
- Fix in recent commit in nvcc_compiler · 1d4f75a7
  由 Pascal Lamblin 提交于 9月 06, 2012
  
  1d4f75a7
- Test img2neibs with only float32 on GPU, other dtypes on CPU · b18629bd
  由 Pascal Lamblin 提交于 9月 06, 2012
  
  b18629bd
- Only tranfer to gpu if dtype is float32 · aa9590f9
  由 Pascal Lamblin 提交于 9月 06, 2012
  
  aa9590f9
- Do not ignore predefined mode in InferShapeTester · 86965f6d
  由 Pascal Lamblin 提交于 9月 06, 2012
  
  86965f6d
- Merge pull request #904 from nouiz/small · 7478227f
  由 lamblin 提交于 9月 06, 2012
```
Small
```
  7478227f
06 9月, 2012 12 次提交
- Merge pull request #906 from pascanur/gpu_opt_scan_outputs_different_dtypes_rebased · edfc726f
  由 nouiz 提交于 9月 06, 2012
```
Gpu opt scan outputs different dtypes rebased
```
  edfc726f
- remove imports to cuda · 553afd8a
  由 Razvan Pascanu 提交于 9月 05, 2012
  
  553afd8a
- remove explicit call to cuda · fcf84d8d
  由 Razvan Pascanu 提交于 9月 05, 2012
  
  fcf84d8d
- add node as an argument to the cython function · 4e78cf82
  由 Razvan Pascanu 提交于 9月 05, 2012
  
  4e78cf82
- bump up the version number · 674c99fc
  由 Razvan Pascanu 提交于 9月 05, 2012
  
  674c99fc
- changes generated by cython · c741009e
  由 Razvan Pascanu 提交于 9月 05, 2012
  
  c741009e
- removed explicit call to cuda.zeros · 1df9dacd
  由 Razvan Pascanu 提交于 9月 05, 2012
  
  1df9dacd
- fixed a few typos · 8d095a6d
  由 Razvan Pascanu 提交于 9月 04, 2012
  
  8d095a6d
- add assertion that dtype is float32 into the function · f0217c61
  由 Razvan Pascanu 提交于 9月 04, 2012
  
  f0217c61
- better error messages · f835b5ac
  由 Razvan Pascanu 提交于 9月 03, 2012
  
  f835b5ac
- new test for mixture dtype outputs of scan · ee776388
  由 Razvan Pascanu 提交于 9月 03, 2012
  
  ee776388
- generate cuda memory only when dtype is float32 · 12ff2b84
  由 Razvan Pascanu 提交于 9月 03, 2012
  
  12ff2b84