stereomatchingkiss
|
4f307dd711
|
implement serialize
|
2016-02-27 23:51:27 +08:00 |
|
stereomatchingkiss
|
277990681b
|
implement serialize
|
2016-02-27 23:45:22 +08:00 |
|
stereomatchingkiss
|
ad598ccccb
|
implement serialize
|
2016-02-27 16:24:02 +08:00 |
|
stereomatchingkiss
|
fb7f43512a
|
implement serialize
|
2016-02-27 16:23:08 +08:00 |
|
stereomatchingkiss
|
2cbd9589ac
|
implement serialize
|
2016-02-27 16:22:31 +08:00 |
|
stereomatchingkiss
|
9b672da91c
|
only serialize weights
|
2016-02-27 15:53:56 +08:00 |
|
stereomatchingkiss
|
f64a84b072
|
implement serialize
|
2016-02-27 15:52:49 +08:00 |
|
stereomatchingkiss
|
7b85147f59
|
implement serialize
|
2016-02-27 15:51:56 +08:00 |
|
stereomatchingkiss
|
2cba975c7e
|
implement serialize
|
2016-02-27 15:49:59 +08:00 |
|
stereomatchingkiss
|
9176684485
|
implement serialize and remove move constructor and move assignemnt since
c++11 should be able to generate them.Remember to specify [ARMA_USE_CXX11]
|
2016-02-27 15:39:48 +08:00 |
|
stereomatchingkiss
|
30d652d6db
|
implement serialize
|
2016-02-27 15:37:15 +08:00 |
|
Ryan Curtin
|
7a77a14684
|
Remove unnecessary checks for NULL.
|
2016-02-25 17:44:15 -05:00 |
|
marcus
|
a215b1b557
|
Test the RMSprop optimizer with logistic regression and a feedforward network.
|
2016-02-25 19:33:08 +01:00 |
|
marcus
|
dc368011ee
|
Minor style fix; Add some spaces.
|
2016-02-24 20:47:42 +01:00 |
|
Ryan Curtin
|
2aedd27420
|
Fix invalid accesses.
|
2016-02-23 12:13:34 -05:00 |
|
Ryan Curtin
|
4a6531bc06
|
Fix invalid accesses.
|
2016-02-23 12:13:06 -05:00 |
|
marcus
|
a1486ebcf5
|
Use testData and testResponses to create the test set.
|
2016-02-23 16:06:35 +01:00 |
|
Ryan Curtin
|
4ce5e6f1d2
|
Fix -Wmaybe-uninitialized.
In this case the warning is wrong, but I don't mind spending what, an extra
cycle when Adaboost starts?, to avoid the warning.
|
2016-02-22 12:06:56 -08:00 |
|
Ryan Curtin
|
f3d692c012
|
Update HISTORY with MiniBatchSGD changes.
|
2016-02-22 10:32:56 -08:00 |
|
Ryan Curtin
|
6e8e87311b
|
Allow mini-batch SGD optimizer.
|
2016-02-22 10:32:56 -08:00 |
|
marcus
|
29dc196be3
|
Widen tolerance slightly.
|
2016-02-22 19:21:53 +01:00 |
|
marcus
|
d765779ba6
|
Set strict = false (use auxiliary memory until the size change) when creating the gradient storage; the default setting changed in armadillo 6.000 from true to false since.
|
2016-02-22 19:18:52 +01:00 |
|
Ryan Curtin
|
4f4bd200d2
|
Add support for MiniBatchSGD.
|
2016-02-22 10:07:48 -08:00 |
|
Ryan Curtin
|
0341d4d82c
|
Add better test for mini-batch SGD, and fix bug.
Also remove debugging output.
|
2016-02-22 09:08:49 -08:00 |
|
Ryan Curtin
|
a85e4c2559
|
Actually use SGD for the test.
|
2016-02-22 09:08:49 -08:00 |
|
marcus
|
efe49b78c2
|
Remove debug output.
|
2016-02-20 16:13:56 +01:00 |
|
marcus
|
384556ec9b
|
Adjust the activation function test; Use the simplified layer structure.
|
2016-02-19 15:53:01 +01:00 |
|
marcus
|
491123bfa2
|
Add SFINAE pattern for the Weights, Reward and Location function.
|
2016-02-19 15:36:02 +01:00 |
|
marcus
|
8b187fcbf3
|
Remove hard tanh activation function header.
|
2016-02-19 15:19:47 +01:00 |
|
marcus
|
6205f3e0b6
|
Update CMakeLists file to build the feedforward, recurrent and convolutional neural network.
|
2016-02-19 15:05:21 +01:00 |
|
marcus
|
53637f2853
|
Add network auxiliary functions test.
|
2016-02-19 15:03:09 +01:00 |
|
marcus
|
0aeb767485
|
Adjust the network test; Use the simplified layer structure.
|
2016-02-19 15:01:37 +01:00 |
|
marcus
|
a49e0feedb
|
Add auxiliary network function to get the size of the network.
|
2016-02-19 14:58:41 +01:00 |
|
marcus
|
cf3c454800
|
Remove the optimizer and weight initialization rule from each layer.
|
2016-02-19 14:54:19 +01:00 |
|
marcus
|
43f0f13fe8
|
Adjust CNN class so that it works with the mlpack optimizers; Add Train() method regarding the design guidelines.
|
2016-02-19 14:34:44 +01:00 |
|
marcus
|
f6c27ed386
|
Adjust RNN class so that it works with the mlpack optimizers; Add Train() method regarding the design guidelines.
|
2016-02-19 14:33:35 +01:00 |
|
marcus
|
69f97e50ae
|
Adjust FFN class so that it works with the mlpack optimizers; Add Train() method regarding the design guidelines.
|
2016-02-19 14:31:43 +01:00 |
|
marcus
|
b2d87c715c
|
Add test for RMSprop.
|
2016-02-19 14:22:53 +01:00 |
|
marcus
|
da1207a9e6
|
Add RMSprop implementation.
|
2016-02-19 14:22:53 +01:00 |
|
marcus
|
1bdfce1f5e
|
Minor spelling fix.
|
2016-02-19 14:22:53 +01:00 |
|
Ryan Curtin
|
f6dd2f7a97
|
Add documentation for change.
|
2016-02-18 11:17:56 -08:00 |
|
Ryan Curtin
|
4e069ab995
|
Handle dimensions with no variance better (i.e. ignore them).
|
2016-02-18 08:37:47 -08:00 |
|
Ryan Curtin
|
50629c1d76
|
Handle MSVC compiler properly; Visual Studio only supports OpenMP 2.5.
OpenMP 3.0 is where unsigned indices are supported.
See also:
https://github.com/VespucciProject/MLPACK_for_MSVC
http://stackoverflow.com/questions/2820621/why-arent-unsigned-openmp-index-variables-allowed
|
2016-02-17 09:03:38 -08:00 |
|
Ryan Curtin
|
c751e61c8e
|
Fix #522, for real this time, by using std::.
|
2016-02-16 22:18:21 -05:00 |
|
Ryan Curtin
|
9c1bcf6d8a
|
Switch to C++ headers instead of old C-style. Fixes #522.
Apparently I never got the memo that I should be using cmath instead of math.h
(and cstdio instead of stdio.h, and so forth).
|
2016-02-16 22:02:32 -05:00 |
|
Ryan Curtin
|
46e0a233c2
|
Add test for MiniBatchSGD.
The second test doesn't work yet, but I think it's because I have set up the
problem wrong.
|
2016-02-12 14:47:23 -08:00 |
|
Ryan Curtin
|
e2acdb7156
|
Remove unnecessary includes.
|
2016-02-12 14:47:08 -08:00 |
|
Ryan Curtin
|
2c476c24c2
|
Add minibatch SGD implementation.
|
2016-02-12 14:46:57 -08:00 |
|
Ryan Curtin
|
d86bccc231
|
Better documentation.
|
2016-02-12 12:49:54 -08:00 |
|
Ryan Curtin
|
ec35c2d4bc
|
Use the correct type.
|
2016-02-12 12:39:50 -08:00 |
|