Commit Graph

143 Commits

Author SHA1 Message Date
ebuehrle
eae8c7f3f4 Fix mask computation 2021-10-19 09:35:14 +02:00
ebuehrle
466e6b6ce7 No-mask options GAIL 2021-10-18 13:28:32 +02:00
Arec
62bfe797a4 adding function to load experts and combine them into single transition list 2021-10-15 02:47:52 -07:00
Arec
ffb16cfc31 updating function to process all expert data from track files, starting processing options policy from file 2021-10-15 02:39:49 -07:00
Arec
415d607418 making the expert demonstration processor go through all agents in order when producing a single (default) trajectory file, using a randomized agent environment in optionsgail, starting function to process and store all expert data 2021-10-13 11:20:40 -07:00
Arec
70e55327dc fixing flataction discriminator to convert to float beforehand, adding necessary forward calls in expert, adding Fire to video creator from model, and trying full run of options gail with new discrimination model 2021-10-11 08:47:18 -07:00
Arec
01752fac12 adding model to append actions to encoded image state for discriminator, files to test different environment experiments, and a script to render an environment from a saved model 2021-10-11 07:14:03 -07:00
Johannes Fischer
244b85a06e Add larger expert dataset creation command 2021-10-08 18:23:59 +02:00
Johannes Fischer
8a67d3536e Add ray directory to gitignore 2021-10-08 18:20:20 +02:00
Johannes Fischer
c59da3611c Use ray tune in gail 2021-10-08 18:19:08 +02:00
Johannes Fischer
6d2ab54b6e Add callback to gail image random to report metrics
Currently nothing is reported yet
2021-10-08 18:18:42 +02:00
Johannes Fischer
4b9a81080b Add missing * in **kwargs 2021-10-08 18:17:02 +02:00
ebuehrle
59681cb16f Merge pull request #3 from sisl/refactor-sampling
Refactor sampling
2021-09-15 09:08:13 +02:00
ebuehrle
4928458e08 Fix discriminator reward
Had wrong sign.
2021-09-15 07:15:43 +02:00
ebuehrle
40da84393c Update imitation version 2021-09-14 18:04:53 +02:00
ebuehrle
183657dc36 Fix buffer bug, add test
Buffer was not being cleared between plan rollouts
2021-09-14 17:55:02 +02:00
ebuehrle
d2932374d9 Refactor sampling 2021-09-14 13:42:25 +02:00
Johannes Fischer
763a7bb0d3 Merge branch 'main' of github.com:sisl/InteractionImitation 2021-09-14 11:01:24 +02:00
ebuehrle
deaef45943 Add test for discriminator 2021-09-14 07:19:58 +02:00
Johannes Fischer
2280597db6 Add GAIL with random agent data 2021-09-13 19:42:41 +02:00
ebuehrle
7eae74a7d8 Switch back to additive reward 2021-09-13 17:55:50 +02:00
johannes-fischer
b0b358544f Merge pull request #2 from sisl/fischer/fusion_sample_methods
Merge different methods to sample the policy and collect transitions
2021-09-13 16:51:16 +02:00
Johannes Fischer
9c7e6cef3a Rename action to option 2021-09-13 16:43:45 +02:00
Johannes Fischer
9b8ceed9c9 Render only one episode 2021-09-13 16:31:24 +02:00
Johannes Fischer
826c0fa219 Merge different methods to sample the policy and collect transitions 2021-09-13 15:50:49 +02:00
ebuehrle
f1ece358d7 Speed up collision check, assume 0 is fallback option
Due to the conservative approximation of the collision check,
no option might be feasible, thus the necessity of a guaranteed fallback.
2021-09-13 09:35:49 +02:00
ebuehrle
87ff3dbb93 Add cuda support, normalize actions 2021-09-11 21:32:30 +02:00
ebuehrle
e7b0aea427 run options gail 2021-09-11 21:32:25 +02:00
Johannes Fischer
8ad7457159 Merge branch 'main' of github.com:sisl/InteractionImitation 2021-09-10 11:38:31 +02:00
Johannes Fischer
544ea4d15a Allow variable horizon trajectories 2021-09-10 11:37:34 +02:00
Johannes Fischer
9a107b165a Allow variable horizon trajectories 2021-09-10 11:34:36 +02:00
Johannes Fischer
b2b2abafa2 Fix typo 2021-09-10 11:34:13 +02:00
ebuehrle
5ff4b42c0e Fix imports 2021-09-08 20:45:45 +02:00
ebuehrle
3a6139286d Update .gitignore 2021-09-08 20:31:45 +02:00
ebuehrle
50916aec05 Vanilla GAIL on rasterized observation 2021-09-08 20:30:11 +02:00
ebuehrle
802d4a4301 Try lower image resolution 2021-09-08 20:26:54 +02:00
ebuehrle
f94ec9a4dc Add test for discriminator 2021-09-08 20:23:39 +02:00
Arec
de5877aaad filling in available_actions, generate_plan, and feasible helpers 2021-09-08 05:14:15 -07:00
ebuehrle
1fd0a71646 Draft Options GAIL 2021-09-08 11:21:25 +02:00
ebuehrle
d89e491b92 Remove debug print statement 2021-09-07 19:20:28 +02:00
ebuehrle
e9f09cacb7 Add option to render expert rollout 2021-09-07 19:19:53 +02:00
ebuehrle
88e0b99d7e Add script for data generation 2021-09-07 19:18:59 +02:00
ebuehrle
a70907c0fd Copy over experiments 2021-09-01 16:07:15 +02:00
Johannes Fischer
317d329765 Add shell script for value dice training 2021-08-06 18:53:32 +02:00
Johannes Fischer
88b4466e57 MInor change in value dice loss, activate print statements, only do EITHER value OR policy update for each batch 2021-08-06 18:52:17 +02:00
Johannes Fischer
025c71767f Change final value network activation to identity 2021-08-06 18:48:53 +02:00
Johannes Fischer
66bfba3986 Minor formatting 2021-08-05 18:37:54 +02:00
Johannes Fischer
6afb112277 Bugfix in value dice
FIRST backward() has to be called on both, policy and value, before step() is called for either of them
2021-08-05 18:35:14 +02:00
Johannes Fischer
bf4c19a4d0 Add value dice ray config 2021-08-05 18:33:34 +02:00
Johannes Fischer
8bce5d15f6 Restore train_epochs to 200 instead of 8 2021-08-05 18:32:46 +02:00