-
7462e69c61
temp 4
temp
Connor Olding
2019-03-11 07:15:41 +0100
-
08f476e6ac
temp 3
Connor Olding
2019-02-26 21:41:37 +0100
-
a1429a6271
temp 2
Connor Olding
2018-08-13 11:58:45 +0200
-
b7938a1785
temp
Connor Olding
2018-06-30 20:08:49 +0200
-
dc8969469d
move param/sigma decay into es methods
master
Connor Olding
2018-06-28 11:03:28 +0200
-
450bd70d99
update presets
Connor Olding
2018-06-28 09:49:01 +0200
-
5c8658312e
make ARS param decay relative to more stuff
Connor Olding
2018-06-24 12:20:34 +0200
-
08148c6736
use normalizing on broadcast tiles
Connor Olding
2018-06-24 12:18:49 +0200
-
18e4376aae
add normalizing and no-biasing features to DenseBroadcast
Connor Olding
2018-06-24 12:18:30 +0200
-
f52fabc549
fix and rewrite ARS telling (it was super broken!)
Connor Olding
2018-06-21 17:13:35 +0200
-
102eefe98c
add PowerSign momentum to ARS, antithetic by default
Connor Olding
2018-06-21 05:14:45 +0200
-
dc235f5d18
add hidden layer settings
Connor Olding
2018-06-17 23:48:43 +0200
-
ebc494fb60
overhaul presets system
Connor Olding
2018-06-17 05:48:15 +0200
-
49dfcfe5b3
scale param decay by sigma (prevents unlearning?)
Connor Olding
2018-06-16 21:46:22 +0200
-
31f153aec0
add video link
Connor Olding
2018-06-16 01:01:24 +0200
-
7a5ba49356
don't turbo in playback mode
Connor Olding
2018-06-16 00:50:22 +0200
-
c3929d8aa1
remove some ancient useless code
Connor Olding
2018-06-16 00:49:05 +0200
-
2fe009b5fe
don't base max_time on number of trials (fixes playback mode)
Connor Olding
2018-06-16 00:44:34 +0200
-
b0058db80e
TODOs and delete an unused variable
Connor Olding
2018-06-16 00:38:09 +0200
-
155f868f56
another attempt at fixing preset argument
Connor Olding
2018-06-16 00:37:19 +0200
-
e3695bfb84
rename weight* to param* outside of nn.lua
Connor Olding
2018-06-16 00:33:11 +0200
-
f3fc95404c
overhaul learning rates:
Connor Olding
2018-06-16 00:24:55 +0200
-
474bac45b8
update xnes2 preset with better sample efficiency
Connor Olding
2018-06-16 00:06:58 +0200
-
655cb6f5b3
add snes2 xnes2 presets, cleanup, tweaks etc.
Connor Olding
2018-06-14 22:41:43 +0200
-
33eeaf9c76
cleanup
Connor Olding
2018-06-14 22:40:52 +0200
-
f512f8ac3a
add sigma decay to xNES
Connor Olding
2018-06-14 22:25:54 +0200
-
422468dd47
more comments
Connor Olding
2018-06-14 22:15:49 +0200
-
63583789c3
use locals; fix fitness_shaping and graycode
Connor Olding
2018-06-13 22:51:12 +0200
-
a1ec797de0
add playable preset
Connor Olding
2018-06-13 22:46:57 +0200
-
1fba61e1b9
one more attempt at fixing the preset argument
Connor Olding
2018-06-13 22:46:31 +0200
-
ac4c534185
fix non-antithetic case for ARS
Connor Olding
2018-06-13 22:46:09 +0200
-
601d78bfda
add evaluation counting to ARS, cleanup
Connor Olding
2018-06-13 21:54:04 +0200
-
6498b4143f
tweak inputs: add power-up status, remove top/bottom tile rows
Connor Olding
2018-06-13 20:18:10 +0200
-
6fa042eda5
fix preset failing to default
Connor Olding
2018-06-13 06:08:32 +0200
-
46f1020bfa
fix xNES telling
Connor Olding
2018-06-13 03:01:17 +0200
-
7800510d1f
add xNES preset, add options, allow preset specified by argument
Connor Olding
2018-06-13 03:00:42 +0200
-
403127bd66
log decisions counter
Connor Olding
2018-06-13 03:00:05 +0200
-
719edfa4ae
remove unused functions from config
Connor Olding
2018-06-13 02:07:14 +0200
-
b8a197eae4
add usage instructions
Connor Olding
2018-06-13 02:03:26 +0200
-
16acc68850
add presets to config, update config in general
Connor Olding
2018-06-13 02:02:56 +0200
-
b4e49d08b9
restore step logging, remove adamant (for now)
Connor Olding
2018-06-13 01:36:40 +0200
-
5c64fcf395
overhaul SNES (importance sampling, adaptation sampling, etc)
Connor Olding
2018-06-13 01:19:32 +0200
-
7bb9c79367
allow Dense layers without biasing
Connor Olding
2018-06-13 01:02:41 +0200
-
fca4779e56
fix training without an unperturbed trial
Connor Olding
2018-06-13 01:00:15 +0200
-
b7a9360d6d
add min_time setting (cap_time -> max_time)
Connor Olding
2018-06-13 00:59:36 +0200
-
7cecd57d05
tweak inits and norm_in for variances of 1
Connor Olding
2018-06-12 23:39:13 +0200
-
74eb2bfbef
fix antithetic ARS
Connor Olding
2018-06-12 20:51:08 +0200
-
ccce6a2d55
sigma tweaks
Connor Olding
2018-06-12 05:39:22 +0200
-
12098ee592
add normalize_sums utility function
Connor Olding
2018-06-12 05:37:55 +0200
-
0d28db0fc4
allow division of input size in Dense layers
Connor Olding
2018-06-12 05:37:35 +0200
-
6c554e0f49
use local in LayerNorm
Connor Olding
2018-06-12 05:36:57 +0200
-
50a7ba78f9
make filenames local to main
Connor Olding
2018-06-12 05:36:24 +0200
-
4a09280be4
add pdf and cdf functions
Connor Olding
2018-06-11 08:11:23 +0200
-
fa0287d966
add sigma decay; move printing to start of epoch
Connor Olding
2018-06-10 19:34:17 +0200
-
56f7c01256
fix network loading
Connor Olding
2018-06-10 19:34:06 +0200
-
bc655979af
display decisions made instead of frame count
Connor Olding
2018-06-10 16:48:02 +0200
-
19cd10382f
use experimental config/network
Connor Olding
2018-06-10 16:41:45 +0200
-
401effbc23
insignificant tweaks
Connor Olding
2018-06-10 16:41:32 +0200
-
70742ccf93
temporarily remove sprite type (it's busted anyway)
Connor Olding
2018-06-10 16:41:07 +0200
-
3eebbc534a
add SNES optimizer
Connor Olding
2018-06-10 16:40:20 +0200
-
d87b8e7118
add mean adaptation hyperparameter
Connor Olding
2018-06-10 16:38:25 +0200
-
47eb173dac
add exists utility function
Connor Olding
2018-06-10 16:35:28 +0200
-
771650613c
move dot_mv to nn
Connor Olding
2018-06-10 16:34:20 +0200
-
0100934ac4
add antithetic sampling for xNES
Connor Olding
2018-06-10 16:33:38 +0200
-
695730335c
set experimental config
Connor Olding
2018-06-09 19:12:25 +0200
-
d6cc49cde1
fix learning without negate_trials
Connor Olding
2018-06-09 18:44:47 +0200
-
bcb6cb9da1
add xNES optimizer
Connor Olding
2018-06-09 18:27:13 +0200
-
fe9494b0d5
refactor ARS out of main (breaks a bunch of stuff)
Connor Olding
2018-06-09 17:56:18 +0200
-
d3e6441c40
reduce tile input to 5 per row using new layers
Connor Olding
2018-06-09 16:20:20 +0200
-
dd5ec3dbde
make network linear
Connor Olding
2018-06-09 16:20:07 +0200
-
2b4bffb401
add Reshape and DenseBroadcast layers
Connor Olding
2018-06-09 16:17:52 +0200
-
ae331ce60b
remove remnants of backwards pass
Connor Olding
2018-06-09 15:23:14 +0200
-
f03e80b1b6
rename notice
Connor Olding
2018-06-09 09:47:33 +0200
-
cbb094adc9
restore flagpole bonus, add missing overlay check
Connor Olding
2018-06-09 04:35:09 +0200
-
9fb98d3fe0
allow setting of world-level, plus random option
Connor Olding
2018-06-09 04:34:21 +0200
-
81d6b509d0
detect when mario is controllable
Connor Olding
2018-06-09 01:43:22 +0200
-
5f85b92b6d
simplify gameconfig button specification
Connor Olding
2018-06-08 23:59:55 +0200
-
9b23327df4
add score multiplier
Connor Olding
2018-06-08 23:59:43 +0200
-
431a591481
fix offscreen sprites sometimes being visible
Connor Olding
2018-06-08 15:03:09 +0200
-
f576a47282
make sprite inputs relative to center of screen
Connor Olding
2018-06-08 14:52:04 +0200
-
fec148fb79
don't turbo in playable mode, note overlay bug
Connor Olding
2018-06-08 14:51:17 +0200
-
c30f07f407
prevent reward gained from fireworks
Connor Olding
2018-06-08 14:12:21 +0200
-
c40e1f929d
fix skipped inputs on lag frames
Connor Olding
2018-06-08 13:48:59 +0200
-
912e114efe
update todo
Connor Olding
2018-06-08 13:47:32 +0200
-
e24c3d31a4
use argsort
Connor Olding
2018-06-08 13:46:38 +0200
-
d33bdfea62
add argsort function
Connor Olding
2018-06-08 02:46:00 +0200
-
374fa4d876
cleanup
Connor Olding
2018-06-08 02:45:07 +0200
-
37d404e77d
reduce embed layer to values actually used ingame
Connor Olding
2018-06-07 22:40:31 +0200
-
9c8c1ccd0c
add tanh activation
Connor Olding
2018-05-14 08:27:11 +0200
-
3030e83d00
refactor learn_from_epoch
Connor Olding
2018-05-14 01:34:08 +0200
-
ec19774af5
localize a couple more things
Connor Olding
2018-05-12 23:08:00 +0200
-
15f0292485
remove defer_prints option (now always true)
Connor Olding
2018-05-12 22:56:04 +0200
-
0fb3b1780f
remove some old comments
Connor Olding
2018-05-12 22:55:04 +0200
-
a836314b8b
refactor game and utility functions
Connor Olding
2018-05-12 22:38:51 +0200
-
7f34de8e7c
add cosine activation
Connor Olding
2018-05-12 21:51:00 +0200
-
7db43038ac
adjust range of timed inputs to stdev of roughly 1
Connor Olding
2018-05-07 16:27:51 +0200
-
7357c8ed62
move LayerNorm after Relu
Connor Olding
2018-05-07 16:22:48 +0200
-
e3a8a6b87f
tweak config
Connor Olding
2018-05-07 16:22:02 +0200
-
946f05bd3e
base timed inputs on start of trial time
Connor Olding
2018-05-07 16:20:59 +0200
-
3e7aeb3c91
config tweaks and fixes
Connor Olding
2018-05-07 09:20:01 +0200