Tue, 28 Jul 2009 11:50:07 -0400 |
Xavier Glorot |
load mecanism no Tie model from Tie model for DAA_input_groups
|
Tue, 28 Jul 2009 10:07:01 -0400 |
Xavier Glorot |
small change (daa input groups)
|
Mon, 27 Jul 2009 19:07:22 -0400 |
Xavier Glorot |
added a supupdate method for DAA inputs groups
|
Thu, 23 Jul 2009 12:25:47 -0400 |
Xavier Glorot |
redifinition of saturation monitoring instance methods DAA_inputs_groups
|
Tue, 21 Jul 2009 17:22:20 -0400 |
Xavier Glorot |
fixed cost scaling bug for DAA_inputs_groups
|
Tue, 21 Jul 2009 13:31:43 -0400 |
Xavier Glorot |
DAA_inputs_groups readability
|
Tue, 21 Jul 2009 12:19:30 -0400 |
Xavier Glorot |
variable scale_cost for DAA_inputs_groups
|
Mon, 20 Jul 2009 13:07:51 -0400 |
Xavier Glorot |
fixed a bug in unsupupdate instance method for DAA_inputs_groups
|
Mon, 20 Jul 2009 13:04:18 -0400 |
Xavier Glorot |
added a separated auxiliary and normal inputs gradient computation behavior
|
Wed, 15 Jul 2009 13:19:56 -0400 |
Frederic Bastien |
put in comment a fct call all that seam to don't be used and conflict with numpy and python name.
|
Wed, 15 Jul 2009 13:18:55 -0400 |
Frederic Bastien |
get the mnist data from the pmat file and not the amat file
|
Mon, 13 Jul 2009 16:56:31 -0400 |
Xavier Glorot |
check saturation and noiseseed methods for DAA input groups
|
Fri, 10 Jul 2009 18:14:36 -0400 |
Xavier Glorot |
added a save and load mecanism and fixed bug to DAA_inputs_groups
|
Fri, 10 Jul 2009 16:56:52 -0400 |
Xavier Glorot |
changed behavior of DAAig and StackedDAAig, added a cost scaling parameter
|
Fri, 10 Jul 2009 12:16:20 -0400 |
Frederic Bastien |
fix FillMissing debug mode error. The error was caused by the fact that debug_mode generate an error by default for NaN value. We must disable this check for this op as we deal with NaN value.
|
Fri, 10 Jul 2009 11:31:32 -0400 |
Frederic Bastien |
gix on gcc4.1 FillMissing, seam broken on gcc 4.3
|
Tue, 30 Jun 2009 12:02:11 -0400 |
phamel |
Reverted previous commit
|
Tue, 30 Jun 2009 11:57:20 -0400 |
phamel |
Test commit - Will be reverted
|
Mon, 29 Jun 2009 16:08:00 -0400 |
Frederic Bastien |
dummy change to force the compilation with the good paramter of FillMissing.c_code()
|
Mon, 29 Jun 2009 11:55:29 -0400 |
desjagui |
* fixed test_stacker.py
|
Mon, 29 Jun 2009 09:49:28 -0400 |
Frederic Bastien |
removed manual call to Member and Variable as this is deprecated in theano.
|
Fri, 26 Jun 2009 16:44:55 -0400 |
Olivier Delalleau |
Merged
|
Fri, 26 Jun 2009 09:55:32 -0400 |
Olivier Delalleau |
Fixed test_float
|
Fri, 26 Jun 2009 10:46:45 -0400 |
Frederic Bastien |
implemented FillMissing.c_code. It use the new c_no_compile_args to remove -ffast-math and -ffinite-math-only as they broke isnan.
|
Wed, 24 Jun 2009 10:53:55 -0400 |
Olivier Delalleau |
Fixed some SGD tests
|
Sun, 21 Jun 2009 20:16:23 -0400 |
Xavier Glorot |
Sigmoid before logistic regression to avoid small gradient with tanh in StackedDAAig
|
Sat, 13 Jun 2009 22:02:13 -0400 |
Foo Bar |
Merge
|
Sat, 13 Jun 2009 10:11:43 -0400 |
Xavier Glorot |
change seed initialisation for random object and weigth init for logistic regression
|
Thu, 11 Jun 2009 15:44:32 -0400 |
Frederic Bastien |
small optimisation
|
Thu, 11 Jun 2009 13:35:31 -0400 |
Frederic Bastien |
test more case in FillMissing and print the time of the op.
|
Thu, 11 Jun 2009 11:37:24 -0400 |
Xavier Glorot |
fix a bug for the weight decay
|
Wed, 10 Jun 2009 13:45:52 -0400 |
Frederic Bastien |
remove something that should not have been commited.
|
Wed, 10 Jun 2009 13:43:27 -0400 |
Frederic Bastien |
added more test for the FillMissing op.
|
Wed, 10 Jun 2009 13:42:56 -0400 |
Frederic Bastien |
opt of the FillMissing op. Now 80-90% faster python implementation.
|
Tue, 09 Jun 2009 21:39:50 -0400 |
Frederic Bastien |
scall optimisation to FillMissing.
|
Tue, 09 Jun 2009 21:37:58 -0400 |
Frederic Bastien |
added test for FillMissing op.
|
Mon, 08 Jun 2009 13:14:09 -0400 |
Xavier Glorot |
small change to Checks of scan_inputs_groups ops
|
Mon, 08 Jun 2009 12:14:14 -0400 |
Frederic Bastien |
very small opt.
|
Fri, 05 Jun 2009 15:27:21 -0400 |
Xavier Glorot |
Better declaration of totallocal_update method in StackedDAAig
|
Fri, 05 Jun 2009 15:11:53 -0400 |
Xavier Glorot |
added a totallocal_update method to StackedDAAig
|
Wed, 03 Jun 2009 14:25:56 -0400 |
Xavier Glorot |
Normalized tanh to be equivalent to sigmoid in DAAig
|
Wed, 03 Jun 2009 13:54:31 -0400 |
Frederic Bastien |
fix typo in comment.
|
Wed, 03 Jun 2009 04:07:30 -0400 |
desjagui |
BUG FIXES !!!!!!!
|
Tue, 02 Jun 2009 22:29:49 -0400 |
Pascal Lamblin |
merge
|
Tue, 02 Jun 2009 22:29:21 -0400 |
Pascal Lamblin |
Finalize new validation split: indices are now randomized again inside
|
Tue, 02 Jun 2009 22:26:29 -0400 |
Olivier Delalleau |
Merged
|
Tue, 02 Jun 2009 22:26:04 -0400 |
Olivier Delalleau |
Fix to the 'by_pair' corruption pattern
|
Tue, 02 Jun 2009 14:40:43 -0400 |
Olivier Delalleau |
Fixed bug introduced in previous change meant to improve numerical stability
|
Tue, 02 Jun 2009 21:15:41 -0400 |
Pascal Lamblin |
merge
|
Tue, 02 Jun 2009 21:15:21 -0400 |
Pascal Lamblin |
Prepare the way for the new train/valid split, based on object instances.
|
Tue, 02 Jun 2009 20:21:35 -0400 |
James Bergstra |
merge
|
Tue, 02 Jun 2009 20:21:21 -0400 |
James Bergstra |
better numeric stability in tile_image
|
Tue, 02 Jun 2009 20:20:48 -0400 |
James Bergstra |
fixed typos in wrap_libsvm
|
Tue, 02 Jun 2009 20:20:03 -0400 |
James Bergstra |
added methods kwarg to sgd module
|
Tue, 02 Jun 2009 20:19:30 -0400 |
James Bergstra |
added doc
|
Tue, 02 Jun 2009 13:51:22 -0400 |
Olivier Delalleau |
Merged
|
Tue, 02 Jun 2009 13:50:43 -0400 |
Olivier Delalleau |
New option 'corruption_pattern' to alter the way inputs are corrupted
|
Tue, 02 Jun 2009 11:38:23 -0400 |
Xavier Glorot |
fixed numerical instability for cross entropy cost calculation of DAAig
|
Tue, 02 Jun 2009 11:31:50 -0400 |
Olivier Delalleau |
Added option to allow one to try and reconstruct originally missing inputs
|
Tue, 02 Jun 2009 10:14:00 -0400 |
Olivier Delalleau |
FillMissing can now impute missing values by an array instead of a single constant
|
Mon, 01 Jun 2009 17:44:56 -0400 |
bergstra |
typos in parameter names
|
Mon, 01 Jun 2009 00:22:35 -0400 |
bergstra |
added hard_time_limit to stopper
|
Sun, 31 May 2009 05:07:55 -0400 |
desjagui |
With float64 and normalize=True, dataset doesn't even fit in memory.
|
Fri, 29 May 2009 10:48:20 -0400 |
Xavier Glorot |
fix a bad declaration of the updates in StackedDAAig
|
Thu, 28 May 2009 12:12:34 -0400 |
Xavier Glorot |
changed name to NLL method to the StackedDAAig
|
Thu, 28 May 2009 12:06:33 -0400 |
Xavier Glorot |
added a NLL method to the StackedDAAig
|
Wed, 27 May 2009 23:15:00 -0400 |
desjagui |
added comments to norb_small
|
Wed, 27 May 2009 21:28:25 -0400 |
desjagui |
Official version of smallNORB, with stereo images of size 96x96
|
Wed, 27 May 2009 15:25:56 -0400 |
Olivier Delalleau |
Fixed the SGD tests
|
Wed, 27 May 2009 14:46:01 -0400 |
Olivier Delalleau |
Fixed some imports, but test_stacker is still not working
|
Wed, 27 May 2009 14:15:35 -0400 |
Olivier Delalleau |
Fixed imports so that more tests pass
|
Wed, 27 May 2009 13:57:10 -0400 |
Olivier Delalleau |
Test test_tzanetakis now passes
|
Wed, 27 May 2009 13:54:13 -0400 |
Olivier Delalleau |
Fixed some imports
|
Wed, 27 May 2009 13:53:50 -0400 |
Olivier Delalleau |
Fixed import so that make_test_datasets can be imported without crashing
|
Wed, 27 May 2009 13:26:33 -0400 |
Olivier Delalleau |
Simplified code now that it is not a unittest anymore
|
Wed, 27 May 2009 13:25:20 -0400 |
Olivier Delalleau |
Renamed test_speed.py into check_speed.py so that nosetests does not run it
|
Wed, 27 May 2009 12:08:18 -0400 |
Olivier Delalleau |
Made test_speed faster when run by nosetests
|
Wed, 27 May 2009 11:48:05 -0400 |
Olivier Delalleau |
Fixed test_speed test
|
Wed, 27 May 2009 11:46:08 -0400 |
Olivier Delalleau |
Added empty __init__.py so that old_dataset can be seen as a module
|
Wed, 27 May 2009 09:59:39 -0400 |
Olivier Delalleau |
Merged
|
Wed, 27 May 2009 09:59:25 -0400 |
Olivier Delalleau |
Replaced debug special code for missing values (-123456) by truly missing (NaN)
|
Tue, 26 May 2009 18:45:09 -0400 |
desjagui |
* when creating the sgd minimizer, the user can ask that step_cost compute other
|
Tue, 26 May 2009 17:39:43 -0400 |
Joseph Turian |
Added more descriptive comments to crossentropy and KL divergence.
|
Mon, 25 May 2009 23:13:56 -0400 |
Olivier Delalleau |
Fixes to the handling of missing values in input
|
Mon, 25 May 2009 23:11:38 -0400 |
Olivier Delalleau |
Some fixes to FillMissing, and added MaskGradient (another Op to move elsewhere later)
|
Mon, 25 May 2009 11:45:37 -0400 |
Olivier Delalleau |
Merged
|
Mon, 25 May 2009 11:43:50 -0400 |
Olivier Delalleau |
Added new option 'ignore_missing' to deal with missing values in input
|
Mon, 25 May 2009 11:42:25 -0400 |
Olivier Delalleau |
New class MaskSelect (should probably be moved to another file)
|
Mon, 25 May 2009 10:23:00 -0400 |
Olivier Delalleau |
New class FillMissing (should probably be moved to another file later)
|
Sun, 24 May 2009 17:16:38 -0400 |
Pascal Lamblin |
Bugfix and improvement to NIPS09Stopper
|
Fri, 22 May 2009 15:14:34 -0400 |
Xavier Glorot |
improved readability DAA_inputs_groups and scan_inputs_groups
|
Fri, 22 May 2009 14:13:38 -0400 |
Olivier Delalleau |
Replaced tabs by 4 blank spaces
|
Fri, 22 May 2009 14:12:17 -0400 |
Olivier Delalleau |
Added a safety assert, just in case
|
Fri, 22 May 2009 14:11:51 -0400 |
Olivier Delalleau |
Fixed minor typo in comment
|
Fri, 22 May 2009 13:43:23 -0400 |
Olivier Delalleau |
Cosmetic changes
|
Fri, 22 May 2009 10:04:04 -0400 |
Olivier Delalleau |
Renamed Alloc into alloc
|