Andre Martins
|
f6de1677e1
|
Implemented safe softmax (but doesn't solve the problem yet, we need log-softmax).
|
2016-09-17 02:42:11 +01:00 |
|
Andre Martins
|
eb57df2a3e
|
Some cleaning.
|
2016-09-17 02:00:03 +01:00 |
|
Andre Martins
|
22bbac3287
|
Eliminated non-deterministic behaviour.
|
2016-09-17 01:55:09 +01:00 |
|
Marcin Junczys-Dowmunt
|
09ce7e6349
|
renamed sgd.h to optimizers.h
|
2016-09-16 23:55:47 +02:00 |
|
Marcin Junczys-Dowmunt
|
6e90198426
|
demonstrate settings for need for safe softmax
|
2016-09-16 23:45:35 +02:00 |
|
Marcin Junczys-Dowmunt
|
c0b676c7c9
|
Tidying up optimizers
|
2016-09-16 23:32:09 +02:00 |
|
Marcin Junczys-Dowmunt
|
70f8277eb1
|
Merge branch 'master' of github.com:emjotde/marian
|
2016-09-16 23:27:05 +02:00 |
|
Marcin Junczys-Dowmunt
|
c46012a6c1
|
Adam optimizer
|
2016-09-16 23:26:54 +02:00 |
|
Hieu Hoang
|
ea66688a13
|
Merge ../Marian
|
2016-09-16 19:56:54 +02:00 |
|
Hieu Hoang
|
40ac84ba73
|
add argmax()
|
2016-09-16 19:56:01 +02:00 |
|
Hieu Hoang
|
948789787d
|
add ArgmaxNodeOp
|
2016-09-16 19:49:59 +02:00 |
|
Andre Martins
|
cbc29a0ab1
|
Added ref for the backprop of the fast softmax.
|
2016-09-16 18:41:26 +01:00 |
|
Andre Martins
|
8f1a7382ec
|
Merged.
|
2016-09-16 18:36:30 +01:00 |
|
Andre Martins
|
c54eaf17d5
|
Made e-d use actual MT data instead of synthetic.
|
2016-09-16 18:27:54 +01:00 |
|
Marcin Junczys-Dowmunt
|
2892fe0feb
|
Merge branch 'master' of github.com:emjotde/marian
|
2016-09-16 19:25:10 +02:00 |
|
Marcin Junczys-Dowmunt
|
c77e589267
|
renamed softmax_fast to softmax
|
2016-09-16 19:24:57 +02:00 |
|
Lane Schwartz
|
015be2bf63
|
Merge branch 'master' of https://github.com/emjotde/Marian
|
2016-09-16 19:21:20 +02:00 |
|
Lane Schwartz
|
016c777d42
|
Don't make doc when make all
|
2016-09-16 19:21:15 +02:00 |
|
Marcin Junczys-Dowmunt
|
268d1e799c
|
Merge branch 'master' of github.com:emjotde/marian
|
2016-09-16 19:03:52 +02:00 |
|
Marcin Junczys-Dowmunt
|
12fa4f40d2
|
removed cudaSetDevice
|
2016-09-16 19:03:45 +02:00 |
|
Marcin Junczys-Dowmunt
|
e37a54712d
|
Update LICENSE.md
|
2016-09-16 18:57:11 +02:00 |
|
Marcin Junczys-Dowmunt
|
3f74962df7
|
Merge branch 'master' of github.com:emjotde/marian
|
2016-09-16 18:34:45 +02:00 |
|
Marcin Junczys-Dowmunt
|
257b621db9
|
fudge factor for adagrad
|
2016-09-16 18:34:38 +02:00 |
|
Lane Schwartz
|
6b7ced99d3
|
Merge branch 'master' of https://github.com/emjotde/Marian
|
2016-09-16 18:28:18 +02:00 |
|
Lane Schwartz
|
5e9cad3169
|
Use CMake to invoke Doxygen
|
2016-09-16 18:28:00 +02:00 |
|
Hieu Hoang
|
c824de0e82
|
merge
|
2016-09-16 18:20:19 +02:00 |
|
Hieu Hoang
|
44785d3108
|
Merge branch 'master' of github.com:emjotde/Marian
|
2016-09-16 18:19:31 +02:00 |
|
Hieu Hoang
|
3c34287549
|
don't call testArgMax()
|
2016-09-16 18:19:22 +02:00 |
|
Hieu Hoang
|
94a846f780
|
don't call testArgMax()
|
2016-09-16 18:17:46 +02:00 |
|
Lane Schwartz
|
6534403830
|
Merge branch 'master' of https://github.com/emjotde/Marian
|
2016-09-16 18:15:11 +02:00 |
|
Hieu Hoang
|
732bb9fa7a
|
merge
|
2016-09-16 18:14:47 +02:00 |
|
Lane Schwartz
|
e897f9d6a5
|
Added Doxyfile
|
2016-09-16 18:14:26 +02:00 |
|
Hieu Hoang
|
5173e9e550
|
move gArgMax to tensor_operators. Write ArgMax() with tensors
|
2016-09-16 18:10:09 +02:00 |
|
Marcin Junczys-Dowmunt
|
09b7e15b05
|
removed cudaSetDevice
|
2016-09-16 17:58:58 +02:00 |
|
Hieu Hoang
|
a4111bf1fe
|
input & output are different
|
2016-09-16 17:55:14 +02:00 |
|
Marcin Junczys-Dowmunt
|
15429dd88f
|
sgd variants
|
2016-09-16 17:43:29 +02:00 |
|
Hieu Hoang
|
7803f44a97
|
Merge ../Marian
|
2016-09-16 17:39:31 +02:00 |
|
Andre Martins
|
1b27accaa0
|
Included embedding layer and graphviz part in the e-d.
|
2016-09-16 16:35:17 +01:00 |
|
Hieu Hoang
|
09c65091f4
|
parallel gArgmax
|
2016-09-16 16:15:15 +01:00 |
|
Andre Martins
|
a5630d2786
|
Merge branch 'master' of https://github.com/emjotde/Marian
|
2016-09-16 16:14:18 +01:00 |
|
Andre Martins
|
5ecd222d82
|
Separated graph computatiom from data in the e-d.
|
2016-09-16 16:14:10 +01:00 |
|
Maximiliana Behnke
|
18018841f3
|
Cosmetic changes to comments in tensor.h
|
2016-09-16 17:10:02 +02:00 |
|
Hieu Hoang
|
4ea1156e99
|
try to use parallelism
|
2016-09-16 17:05:51 +02:00 |
|
Maximiliana Behnke
|
b6c25a8db7
|
Add Doxygen comments to tensor.h
|
2016-09-16 16:49:05 +02:00 |
|
Hieu Hoang
|
269b94cea6
|
serial gArgmax
|
2016-09-16 15:40:20 +01:00 |
|
Hieu Hoang
|
32a5985500
|
move temp() to test.cu
|
2016-09-16 16:03:42 +02:00 |
|
Hieu Hoang
|
78444bc2eb
|
can call gSoftmax
|
2016-09-16 14:50:14 +01:00 |
|
Andre Martins
|
c7a1542b02
|
Separating graph from data.
|
2016-09-16 14:25:06 +01:00 |
|
Hieu Hoang
|
b6e4b8f3f3
|
temp()
|
2016-09-16 15:08:09 +02:00 |
|
Hieu Hoang
|
3b0ea609e4
|
temp()
|
2016-09-16 15:02:30 +02:00 |
|