Arec Jamgochian
|
7814c7c30d
|
ignoring pngs
|
2022-02-28 00:37:30 -08:00 |
|
Arec Jamgochian
|
5533faab40
|
adding metric pkls
|
2022-02-28 00:36:52 -08:00 |
|
Arec Jamgochian
|
1ee3805aba
|
adding trained test policies, as well as fixing evaluate script and commands
|
2022-02-27 23:51:12 -08:00 |
|
Arec Jamgochian
|
e37911eeff
|
removing gail-trpo since performance is about the same as gail, adding experiment evaluation script, updating metric averaging to work
|
2022-02-27 23:03:28 -08:00 |
|
Arec Jamgochian
|
e110570092
|
adding final bc configs, updating experiment to validate in longer intervals
|
2022-02-27 20:40:24 -08:00 |
|
Arec Jamgochian
|
5f77398fd0
|
commiting json files and intermediate settings used to run baseline methods around 02-26, 16ocklock. Next will set off big tuning runs for gail, gailtrpo, and bc (with committed values)
|
2022-02-27 16:27:01 -08:00 |
|
Arec Jamgochian
|
2cacdf691f
|
Merge branch 'main' of https://github.com/sisl/InteractionImitation into main
|
2022-02-27 14:30:32 -08:00 |
|
Arec Jamgochian
|
e52b658aff
|
adding all bestconfigs from big tune run (not sure what happened in last commit)
|
2022-02-27 14:29:49 -08:00 |
|
Arec Jamgochian
|
578459e37b
|
best_configs/hail_expA.json
|
2022-02-27 14:27:29 -08:00 |
|
ebuehrle
|
25e4d5679f
|
Make envs independent in BC experiment
|
2022-02-27 22:53:12 +01:00 |
|
Arec Jamgochian
|
7f1557144c
|
committing etiennes best configs, leading to 73% success rate in shail expB
|
2022-02-27 13:43:59 -08:00 |
|
ebuehrle
|
7039468e46
|
Add tuning script for BC. Todo: if enough GPUs, enable resources in run function
|
2022-02-27 17:30:33 +01:00 |
|
ebuehrle
|
cc3848e8e0
|
Remove unused parameters
|
2022-02-27 15:02:32 +01:00 |
|
ebuehrle
|
4443492d3e
|
Add tuning script for GAIL (TRPO)
|
2022-02-27 14:38:40 +01:00 |
|
ebuehrle
|
4b4fa7e09a
|
Add tuning script for GAIL (PPO)
|
2022-02-27 14:14:08 +01:00 |
|
Arec Jamgochian
|
569e0756ca
|
getting loading of configs, overwriting with seeds, running with tune, and moving back to directory working. adding check for either training or testing, and allowing specification of number of test cpus to split seeds over
|
2022-02-27 02:25:16 -08:00 |
|
Arec Jamgochian
|
21cbd1c956
|
getting best shail/hail configs outsave up and running, adding shell for test runs
|
2022-02-27 01:22:39 -08:00 |
|
Arec Jamgochian
|
acb4fdc518
|
renaming shail policy file, removing ogail file since its unnecessary
|
2022-02-27 00:00:42 -08:00 |
|
Arec Jamgochian
|
3280041efa
|
committing what is hopefully final run of sgail for both experiments, A and B
|
2022-02-26 23:48:58 -08:00 |
|
Arec Jamgochian
|
3b9051505e
|
naming convention doesnt like env key, leakyrelu is no bueno, running quick experiment to see which option sets can work with our time left. the options that look 4s ahead are definitely out of the questions, possibly the 2s ones aswell
|
2022-02-26 16:02:29 -08:00 |
|
ebuehrle
|
71e3c5f816
|
Parametrize discriminator architecture
|
2022-02-26 14:53:29 +01:00 |
|
ebuehrle
|
35e6fb299c
|
Use collision rate as main metric
|
2022-02-26 14:28:58 +01:00 |
|
ebuehrle
|
1e24612347
|
Merge branch 'main' of https://github.com/sisl/InteractionImitation
|
2022-02-26 14:06:22 +01:00 |
|
ebuehrle
|
6876cf9625
|
Track collision rate
|
2022-02-26 13:50:53 +01:00 |
|
ebuehrle
|
dc9cbf329b
|
Enable safe options on evaluation env
|
2022-02-26 12:24:25 +01:00 |
|
ebuehrle
|
f9d3cceed5
|
Save checkpoints and config to folder, move params to config
|
2022-02-26 12:23:42 +01:00 |
|
Arec Jamgochian
|
c28c6c05b7
|
added ogail script and splitting up feasability on next line to avoid calculating it unnecessarily (though it might be fine as is)
|
2022-02-25 18:39:33 -08:00 |
|
Arec Jamgochian
|
fa98601fa6
|
fixing issues with lazylinear sequential, setting off a big run
|
2022-02-25 17:48:56 -08:00 |
|
Arec Jamgochian
|
2dfd7e3c2b
|
add policy saving
|
2022-02-25 17:00:46 -08:00 |
|
Arec Jamgochian
|
59083ebce3
|
making option list indexable to visualize in tensorboard. noticing run from last night had much better performance under more long options, unclear if due to choice or environment factors, like episodes lasting longer. making stoponcollision and seed parameters. fixing ability to see reward metrics under ray/tune/, problem was they were being returned as tensors, not floats
|
2022-02-25 16:36:14 -08:00 |
|
Arec Jamgochian
|
d1f23e6d25
|
readding expertdata to gitignore, just manually adding expert joint state and action files
|
2022-02-25 01:27:02 -08:00 |
|
Arec Jamgochian
|
46564231ef
|
changing gitignore to upload expert data for evaluation, adding multiple roundabout trackfile environments in vecenv
|
2022-02-25 01:21:16 -08:00 |
|
Arec Jamgochian
|
bf9ce84fe4
|
changing when directory name gets saved
|
2022-02-25 00:42:44 -08:00 |
|
Arec Jamgochian
|
99aa50a08b
|
added support for different activations, number of hidden layers, options, running a big run over this
|
2022-02-25 00:00:04 -08:00 |
|
Arec Jamgochian
|
7feea74eb8
|
adding stablebaselines, adding safe loading for nonCuda cluster
|
2022-02-24 23:18:02 -08:00 |
|
ebuehrle
|
5a5d8a7aff
|
WIP: Support different plan lengths
|
2022-02-25 01:32:54 +01:00 |
|
ebuehrle
|
02c1813b00
|
Incrementing agent expert data, smaller policy network
|
2022-02-25 01:23:15 +01:00 |
|
Johannes Fischer
|
0d193d4af3
|
Scratch for horner scheme
|
2022-02-24 15:12:29 +01:00 |
|
ebuehrle
|
febceed651
|
Add expert data
|
2022-02-23 18:17:30 +01:00 |
|
ebuehrle
|
336cf02278
|
Parameterize hidden layer size of policy, add some candidates to grid search
|
2022-02-23 18:14:23 +01:00 |
|
ebuehrle
|
7f64ec7bb0
|
Move hyperparameters to config object
ToDo: parameterize network architectures
|
2022-02-23 18:00:19 +01:00 |
|
ebuehrle
|
406c4ad9ee
|
Fix tune by moving file
|
2022-02-23 17:41:41 +01:00 |
|
ebuehrle
|
f037c119cc
|
Set up for ray tune
|
2022-02-23 16:55:13 +01:00 |
|
ebuehrle
|
91f88983b0
|
Add learning rate schedule to SHAIL-PPO
|
2022-02-23 14:22:02 +01:00 |
|
Arec
|
68b066ec53
|
adding success rate, total distance, and survive time
|
2022-02-22 00:48:35 -08:00 |
|
Arec
|
e7f4f6a871
|
wrapping all environments in timelimit to stop runs longer than 100s, since some others were erroring
|
2022-02-21 17:54:29 -08:00 |
|
Arec
|
e7f8385628
|
updating rwse to work at different times, updating correct testing environment from roundabout, removing the assertion that a collision implies done in the evaluator, using nanmean and nanstd in averaging
|
2022-02-21 15:55:16 -08:00 |
|
ebuehrle
|
daa4825f17
|
Add GAIL-PPO
|
2022-02-21 22:21:13 +01:00 |
|
ebuehrle
|
a89317e1d5
|
Correct checkpoint file path
|
2022-02-21 22:11:34 +01:00 |
|
ebuehrle
|
b89f5db9d8
|
Add SHAIL policy with more options
|
2022-02-21 22:07:44 +01:00 |
|