Online Taylor Residual World Model / results.json
Beats tuned baseline
1{
2 "seed": 3100,
3 "rls_exactness": {
4 "max_prefix_coef_error": 7.333214858684656e-06,
5 "final_prediction_error": 5.4545452599086275e-09
6 },
7 "forgetting_tradeoff": {
8 "lambda_.90_recovery_step": 49,
9 "lambda_.99_recovery_step": 150,
10 "lambda_.90_post_std": 0.015350422357626596,
11 "lambda_.99_post_std": 0.0024082248977793956
12 },
13 "dynamics": {
14 "0": {
15 "mean_abs_one_step_error": 0.20458202460413216,
16 "early_postshift_error": 0.2618333007298551,
17 "mean_abs_10_step_rollout_error": 0.6165080623049105
18 },
19 "1": {
20 "mean_abs_one_step_error": 0.049547163713645745,
21 "early_postshift_error": 0.0654819755070897,
22 "mean_abs_10_step_rollout_error": 0.14357635738274085
23 },
24 "2": {
25 "mean_abs_one_step_error": 0.033109660045941985,
26 "early_postshift_error": 0.04276705428537927,
27 "mean_abs_10_step_rollout_error": 0.1103410053125612
28 },
29 "3": {
30 "mean_abs_one_step_error": 0.02296742322528179,
31 "early_postshift_error": 0.02822090600889082,
32 "mean_abs_10_step_rollout_error": 0.07698101873565563
33 }
34 }
35}