Commit Graph

140 Commits

Author SHA1 Message Date
Connor Olding dc8969469d move param/sigma decay into es methods 2018-06-28 11:03:28 +02:00
Connor Olding 450bd70d99 update presets
too many changes to list (or even tell what's changed),
but most importantly, ARS has improved.
2018-06-28 09:49:01 +02:00
Connor Olding 5c8658312e make ARS param decay relative to more stuff
to be consistent with the other optimizers
2018-06-24 12:20:34 +02:00
Connor Olding 08148c6736 use normalizing on broadcast tiles 2018-06-24 12:18:49 +02:00
Connor Olding 18e4376aae add normalizing and no-biasing features to DenseBroadcast 2018-06-24 12:18:30 +02:00
Connor Olding f52fabc549 fix and rewrite ARS telling (it was super broken!) 2018-06-21 17:13:35 +02:00
Connor Olding 102eefe98c add PowerSign momentum to ARS, antithetic by default 2018-06-21 05:15:05 +02:00
Connor Olding dc235f5d18 add hidden layer settings 2018-06-17 23:48:43 +02:00
Connor Olding ebc494fb60 overhaul presets system 2018-06-17 05:52:24 +02:00
Connor Olding 49dfcfe5b3 scale param decay by sigma (prevents unlearning?) 2018-06-16 22:47:57 +02:00
Connor Olding 31f153aec0 add video link 2018-06-16 01:01:24 +02:00
Connor Olding 7a5ba49356 don't turbo in playback mode 2018-06-16 00:55:41 +02:00
Connor Olding c3929d8aa1 remove some ancient useless code 2018-06-16 00:55:41 +02:00
Connor Olding 2fe009b5fe don't base max_time on number of trials (fixes playback mode) 2018-06-16 00:55:41 +02:00
Connor Olding b0058db80e TODOs and delete an unused variable 2018-06-16 00:38:09 +02:00
Connor Olding 155f868f56 another attempt at fixing preset argument 2018-06-16 00:37:19 +02:00
Connor Olding e3695bfb84 rename weight* to param* outside of nn.lua 2018-06-16 00:33:47 +02:00
Connor Olding f3fc95404c overhaul learning rates:
- rename mean_adapt to weight_rate
- sigma and covar update rates can be specified separately
  (sigma_rate, covar_rate)
- base decays on current rates instead of initially configured rates
  (this might break stuff)
- base_rate takes the place of learning_rate
2018-06-16 00:29:15 +02:00
Connor Olding 474bac45b8 update xnes2 preset with better sample efficiency 2018-06-16 00:06:58 +02:00
Connor Olding 655cb6f5b3 add snes2 xnes2 presets, cleanup, tweaks etc. 2018-06-14 22:42:51 +02:00
Connor Olding 33eeaf9c76 cleanup 2018-06-14 22:40:52 +02:00
Connor Olding f512f8ac3a add sigma decay to xNES 2018-06-14 22:40:39 +02:00
Connor Olding 422468dd47 more comments 2018-06-14 22:15:49 +02:00
Connor Olding 63583789c3 use locals; fix fitness_shaping and graycode 2018-06-13 22:52:37 +02:00
Connor Olding a1ec797de0 add playable preset 2018-06-13 22:46:57 +02:00
Connor Olding 1fba61e1b9 one more attempt at fixing the preset argument 2018-06-13 22:46:31 +02:00
Connor Olding ac4c534185 fix non-antithetic case for ARS 2018-06-13 22:46:09 +02:00
Connor Olding 601d78bfda add evaluation counting to ARS, cleanup 2018-06-13 21:54:04 +02:00
Connor Olding 6498b4143f tweak inputs: add power-up status, remove top/bottom tile rows 2018-06-13 20:18:10 +02:00
Connor Olding 6fa042eda5 fix preset failing to default 2018-06-13 06:08:32 +02:00
Connor Olding 46f1020bfa fix xNES telling 2018-06-13 03:01:54 +02:00
Connor Olding 7800510d1f add xNES preset, add options, allow preset specified by argument 2018-06-13 03:01:54 +02:00
Connor Olding 403127bd66 log decisions counter 2018-06-13 03:00:05 +02:00
Connor Olding 719edfa4ae remove unused functions from config 2018-06-13 02:07:14 +02:00
Connor Olding b8a197eae4 add usage instructions 2018-06-13 02:03:26 +02:00
Connor Olding 16acc68850 add presets to config, update config in general 2018-06-13 02:02:56 +02:00
Connor Olding b4e49d08b9 restore step logging, remove adamant (for now) 2018-06-13 01:42:36 +02:00
Connor Olding 5c64fcf395 overhaul SNES (importance sampling, adaptation sampling, etc) 2018-06-13 01:19:32 +02:00
Connor Olding 7bb9c79367 allow Dense layers without biasing 2018-06-13 01:02:41 +02:00
Connor Olding fca4779e56 fix training without an unperturbed trial 2018-06-13 01:00:15 +02:00
Connor Olding b7a9360d6d add min_time setting (cap_time -> max_time) 2018-06-13 00:59:36 +02:00
Connor Olding 7cecd57d05 tweak inits and norm_in for variances of 1 2018-06-12 23:39:13 +02:00
Connor Olding 74eb2bfbef fix antithetic ARS 2018-06-12 20:57:31 +02:00
Connor Olding ccce6a2d55 sigma tweaks 2018-06-12 05:39:22 +02:00
Connor Olding 12098ee592 add normalize_sums utility function 2018-06-12 05:37:55 +02:00
Connor Olding 0d28db0fc4 allow division of input size in Dense layers 2018-06-12 05:37:35 +02:00
Connor Olding 6c554e0f49 use local in LayerNorm 2018-06-12 05:36:57 +02:00
Connor Olding 50a7ba78f9 make filenames local to main 2018-06-12 05:36:24 +02:00
Connor Olding 4a09280be4 add pdf and cdf functions 2018-06-11 08:11:23 +02:00
Connor Olding fa0287d966 add sigma decay; move printing to start of epoch 2018-06-10 19:34:17 +02:00