提交 · 5f8f1fa907774ace92d1f301aecab14871d4e1f1 · testgroup / pytensor

29 8月, 2009 6 次提交
- merge, backport patch about complex. · 5f8f1fa9
  由 Frederic Bastien 提交于 8月 28, 2009
  
  5f8f1fa9
- merge · 0ac475d2
  由 Frederic Bastien 提交于 8月 28, 2009
  
  0ac475d2
- merge · 85ec1f68
  由 Frederic Bastien 提交于 8月 28, 2009
  
  85ec1f68
- merge. · 914a3d10
  由 Frederic Bastien 提交于 8月 28, 2009
  
  914a3d10
- merge · 67323e7f
  由 Frederic Bastien 提交于 8月 28, 2009
  
  67323e7f
- backport from main theano. · 95ded5b4
  由 Frederic Bastien 提交于 8月 28, 2009
  
  95ded5b4
28 8月, 2009 1 次提交
- add the flops of ConvOp in full mode. · 4bba8d3a
  由 Frederic Bastien 提交于 8月 28, 2009
  
  4bba8d3a
21 8月, 2009 1 次提交
- removed change that should never have been their. · cc5593fe
  由 Frederic Bastien 提交于 8月 20, 2009
  
  cc5593fe
20 8月, 2009 11 次提交
- removed an accidental print to stderr · 117d591c
  由 James Bergstra 提交于 8月 19, 2009
  
  117d591c
- made test of crossentropy optimizations more robust · 7dc4ad86
  由 James Bergstra 提交于 8月 19, 2009
  
  7dc4ad86
- fix test: now zeros_like and ones_like return the same type as the model · de39392d
  由 Frederic Bastien 提交于 8月 19, 2009
  
  de39392d
- relaxed the requirement in raw_random that node inputs have the same dtype as op default arguments · 3d01f413
  由 James Bergstra 提交于 8月 19, 2009
  
  3d01f413
- merge · ad942afa
  由 James Bergstra 提交于 8月 19, 2009
  
  ad942afa
- adjusted test of softmax optimization to account for the less-aggressive dimshuffle lift we use now · e3eaaef1
  由 James Bergstra 提交于 8月 19, 2009
  
  e3eaaef1
- fix missing import in test_blas.py · c39dc2b5
  由 Frederic Bastien 提交于 8月 19, 2009
  
  c39dc2b5
- merge · 58f6f6b1
  由 Frederic Bastien 提交于 8月 19, 2009
  
  58f6f6b1
- merge of changeset af0638d6d9bb. Back-port to python 2.4 · d325be92
  由 Frederic Bastien 提交于 8月 19, 2009
  
  d325be92
- merge · 41b7bcc9
  由 James Bergstra 提交于 8月 19, 2009
  
  41b7bcc9
- removed REAL from nnet.py, using dtype_%(varname)s approach instead · c1ccd518
  由 James Bergstra 提交于 8月 19, 2009
  
  c1ccd518
19 8月, 2009 12 次提交
- merge with changeset d3e7a9b34c9c · ba1903f0
  由 Frederic Bastien 提交于 8月 19, 2009
  
  ba1903f0
- merge changeset 9b9f8dd051ed · 041e23b1
  由 Frederic Bastien 提交于 8月 19, 2009
  
  041e23b1
- merge · 3cf18963
  由 Frederic Bastien 提交于 8月 19, 2009
  
  3cf18963
- merge of changeset d2076eaadf2d. · e962a84d
  由 Frederic Bastien 提交于 8月 19, 2009
  
  e962a84d
- added as_tensor_variable to DimShuffle.make_node · c3f2fa12
  由 James Bergstra 提交于 8月 18, 2009
  
  c3f2fa12
- argmax output dtype is now int32 · dd41a16d
  由 James Bergstra 提交于 8月 18, 2009
  
  dd41a16d
- when compilation fails, the source is printed with line numbers to stderr · 4dfd1f0a
  由 James Bergstra 提交于 8月 18, 2009
  
  4dfd1f0a
- set default log level in debugmode back to warning · 57569e64
  由 James Bergstra 提交于 8月 18, 2009
  
  57569e64
- code cleanup. · 8480958c
  由 Frederic Bastien 提交于 8月 18, 2009
  
  8480958c
- added fct ConvOp.set_flops() to work with the profilemode hack · 31179514
  由 Frederic Bastien 提交于 8月 18, 2009
  
  31179514
- implemented a hidden hack in profilemode to print the number of MFlops. Need… · e07d96fc
  由 Frederic Bastien 提交于 8月 18, 2009
```
implemented a hidden hack in profilemode to print the number of MFlops. Need manual intervention and good knowledge of the theano internal to make it work correctly.
```
  e07d96fc
- better alignement of % of time. · bd979bd5
  由 Frederic Bastien 提交于 8月 18, 2009
  
  bd979bd5
18 8月, 2009 1 次提交
- small optimisation in ConvOp. When the unroll value is not a good multiple, we… · 29b40a49
  由 Frederic Bastien 提交于 8月 17, 2009
```
small optimisation in ConvOp. When the unroll value is not a good multiple, we try to find the best one under the gived value.
```
  29b40a49
15 8月, 2009 1 次提交
- ProfileMode now profile the compilation time too. · cd8bc0b0
  由 Frederic Bastien 提交于 8月 14, 2009
  
  cd8bc0b0
14 8月, 2009 4 次提交
- more fix to complex type cast and better test. · a6ea5b1e
  由 Frederic Bastien 提交于 8月 13, 2009
  
  a6ea5b1e
- fixed the fix about complex type and added test for it. Downcast not allow for now. · af43180b
  由 Frederic Bastien 提交于 8月 13, 2009
  
  af43180b
- added complexs, complexs64 and complexs128 that act as floats. · 7ce5a2af
  由 Frederic Bastien 提交于 8月 13, 2009
  
  7ce5a2af
- correctly fix the bug discovered by Joseph and added test case. · f9604d57
  由 Frederic Bastien 提交于 8月 13, 2009
  
  f9604d57
12 8月, 2009 1 次提交
- implement casting from float32(64) to complex64(128). · 39500b2b
  由 Frederic Bastien 提交于 8月 11, 2009
  
  39500b2b
08 8月, 2009 1 次提交
- many changes motivated by theano_cuda_ndarray · 4fd78082
  由 James Bergstra 提交于 8月 07, 2009
  
  4fd78082
07 8月, 2009 1 次提交
- CAReduce output views input when reducing scalars · 3c7f8139
  由 Joseph Turian 提交于 8月 06, 2009
```
See ticket #417: http://www.pylearn.org/theano/trac/ticket/417
```
  3c7f8139