Datasets:

villekuosmanen
/

dAgger_build_block_tower_1.3.0-advantages

frame_index int64 0 56.1k	advantage float64 -1,115.56 1.6k	required_intervention bool 1 class
0	-24.627913	false
1	-20.927396	false
2	-12.067675	false
3	-19.330513	false
4	-10.508821	false
5	-14.334858	false
6	-15.972628	false
7	-19.223768	false
8	0.12515	false
9	-13.612139	false
10	-0.464519	false
11	5.073342	false
12	3.844729	false
13	6.703296	false
14	13.625504	false
15	9.756006	false
16	7.445363	false
17	6.65812	false
18	-0.322606	false
19	0.422034	false
20	7.990143	false
21	4.179069	false
22	6.530027	false
23	9.486857	false
24	-10.275888	false
25	12.103724	false
26	2.896702	false
27	-3.080529	false
28	11.179238	false
29	8.572664	false
30	6.777065	false
31	5.133576	false
32	-0.18765	false
33	-0.271807	false
34	17.445156	false
35	11.614509	false
36	12.292433	false
37	10.880067	false
38	9.934136	false
39	4.87615	false
40	27.573328	false
41	28.315204	false
42	7.135803	false
43	5.443327	false
44	17.574012	false
45	13.531816	false
46	27.643856	false
47	31.818624	false
48	1.314687	false
49	1.160765	false
50	4.435161	false
51	4.655703	false
52	-20.179134	false
53	-8.90336	false
54	-10.873182	false
55	-1.376615	false
56	-16.880244	false
57	-10.739847	false
58	-14.701031	false
59	-19.386554	false
60	-20.466487	false
61	-26.100514	false
62	-26.801598	false
63	-27.469325	false
64	-16.116257	false
65	-19.125983	false
66	-25.034305	false
67	-36.525475	false
68	-2.071695	false
69	-2.454547	false
70	6.383062	false
71	15.587511	false
72	10.279341	false
73	3.396305	false
74	6.367527	false
75	-15.907342	false
76	15.157101	false
77	-3.841657	false
78	8.197628	false
79	7.936675	false
80	23.538947	false
81	32.126278	false
82	12.337609	false
83	25.558902	false
84	2.424166	false
85	11.901576	false
86	2.249562	false
87	0.337496	false
88	-8.153574	false
89	3.823284	false
90	-2.669942	false
91	-6.555547	false
92	-15.216074	false
93	-3.462713	false
94	-31.275634	false
95	-28.707755	false
96	-36.341435	false
97	-38.115591	false
98	2.406152	false
99	8.576858	false

End of preview. Expand in Data Studio

Advantage Values for villekuosmanen/dAgger_build_block_tower_1.3.0

Pre-computed advantage values for offline RL training.

Source

Dataset: villekuosmanen/dAgger_build_block_tower_1.3.0
Value Model: villekuosmanen/rewact_build_block_tower_all_3
N-step lookahead: 50

Files

This dataset contains per-episode parquet files with advantage values for each frame.

Usage

from pathlib import Path
import pandas as pd

# Load advantages for a specific episode
advantage_df = pd.read_parquet("episode_00000.parquet")

Downloads last month: 27