Commit graph

180 commits

Author SHA1 Message Date
a685db1489 add exact GELU activation 2018-03-22 12:02:17 +01:00
7161f983ab fix __name__ being incorrect due to extra __all__
this fixes tracebacks and checks for __main__,
among other things.
2018-03-17 14:09:15 +01:00
ec71d4956e fix module leaking for real 2018-03-13 03:27:11 +01:00
843524b80a move ActivityRegularizer to regularizer.py 2018-03-13 02:34:44 +01:00
b74e0941dc fix Swish inits 2018-03-12 02:17:28 +01:00
bd1e80b8de activation layers inherit a dummy class 2018-03-11 22:34:46 +01:00
cb4e6d3725 add Swish and SiLU activations 2018-03-11 03:34:00 +01:00
91cdea3b26 fix inequalities in HardClip
oldest trick in the book
2018-03-10 05:03:26 +01:00
a6519f5455 improve notes on dependencies 2018-03-09 10:17:31 +01:00
bd4f2a9478 fix missing import for ActivityRegularizer 2018-03-09 10:09:50 +01:00
9a45b26b7f add rough stratified k-folding utility class 2018-03-08 02:41:45 +01:00
65bc9b8a6f optionally allow gradients to passthru Input layer 2018-03-08 02:40:56 +01:00
4746103978 add HardClip activation layer 2018-03-08 02:40:42 +01:00
44cae4ad50 add LookupLearner 2018-03-07 01:58:17 +01:00
8abbb1e713 add NoiseInjector and NoiseMultiplier layers 2018-03-07 01:54:48 +01:00
604ffb9fa1 add variant of L1L2 regularization using averages 2018-03-07 01:53:40 +01:00
713fd2adbe add experimental soft-clipped optimizers 2018-03-07 01:52:26 +01:00
3aa3b70a9f add AMSgrad optimizer 2018-03-07 01:30:04 +01:00
0641c747c9 add Arcsinh activation 2018-03-07 01:29:48 +01:00
8ce2ec1ad4 add missing import 2018-02-10 11:28:43 +01:00
39bbf27860 add onehot utility function 2018-02-02 08:52:32 +01:00
e7783188bb tweak float exports 2018-02-02 08:51:39 +01:00
169303813d basic PEP 8 compliance
rip readability
2018-01-22 19:40:36 +00:00
c81ce0afbb rename stuff and add a couple missing imports 2018-01-21 22:16:36 +00:00
bbdb91fcb1 merge and split modules into a package 2018-01-21 22:07:57 +00:00
db65fbdd62 add Neumann optimizer 2018-01-12 15:42:04 +00:00
1ebb897f14 use @ operator 2017-10-19 04:12:16 +00:00
a85ee67780 allow CLRs to use optimizer's learning rate 2017-10-19 04:03:44 +00:00
763246df98 add RMSpropCentered to model from config 2017-09-26 23:12:40 +00:00
9bb26b1ec5 add Huber loss 2017-09-25 16:37:52 +00:00
eb16377ba6 add Adagrad optimizer 2017-09-25 16:06:45 +00:00
c964f143d2 not true 2017-09-25 07:12:19 +00:00
5b6fd6259f update example 2017-09-25 06:28:59 +00:00
a760c4841b add fallback to optim.lr in AnnealingLearner 2017-09-25 06:10:54 +00:00
916c6fe1f0 assert that rituals have been prepared 2017-09-25 06:10:04 +00:00
615f43c550 support AddSign and PowerSign in config 2017-09-25 06:09:36 +00:00
fe54002671 remove DumbLearner 2017-09-25 06:09:07 +00:00
9a7ffe5f0d add AddSign and PowerSign optimizers 2017-09-25 04:02:17 +00:00
5c83f063be remove keras stuff so it won't show on google
bleh
2017-09-18 04:42:41 +00:00
d38e2076f0 allow multi-input and multi-output models 2017-09-16 18:28:05 +00:00
3386869b30 move actreg tweaking into if statement
i was getting division by zero.
2017-09-16 17:33:47 +00:00
65ba80bb96 skip over irrelevant nodes 2017-09-16 17:31:39 +00:00
dcbaef3032 use MomentumClip in warmup for stability 2017-09-16 17:30:52 +00:00
7878f94f43 auto-increment loss filenames to stop clobbering 2017-09-16 17:30:02 +00:00
e22316a4c9 move losses into Model and refactor methods 2017-09-16 17:05:25 +00:00
910facf98d move NLL to core 2017-08-05 10:59:05 +00:00
0b9c1fe117 allow SGDR to anneal optimizer's learning rate
e.g. YellowFin
2017-08-05 10:43:38 +00:00
dbd6c31ea5 fix final rate calculation 2017-08-05 10:43:18 +00:00
915b39d783 allow Optimizers to inspect Models (currently unused)
the thing that takes advantage of this may or may not be committed,
so this may or may not get reverted.
2017-08-05 10:41:35 +00:00
de5af4f7f4 allow argument passthru to normalizer in _mr_make_norm 2017-08-05 10:40:39 +00:00