Trajectories









Bookmarks

Lowest advantage
episodes
(unexpected failures):

trajectory 7, frame 148
trajectory 8, frame 445
trajectory 6, frame 389
trajectory 6, frame 252
trajectory 1, frame 95
trajectory 4, frame 430
trajectory 2, frame 1
trajectory 1, frame 346
trajectory 1, frame 240
trajectory 8, frame 98
trajectory 2, frame 143
trajectory 5, frame 274
trajectory 2, frame 74
trajectory 8, frame 361
trajectory 7, frame 232
trajectory 3, frame 116

Highest advantage
episodes
(unexpected successes):

trajectory 6, frame 332
trajectory 2, frame 42
trajectory 8, frame 282
trajectory 5, frame 380
trajectory 5, frame 219
trajectory 2, frame 483
trajectory 1, frame 177
trajectory 3, frame 456
trajectory 5, frame 67
trajectory 5, frame 166
trajectory 1, frame 452
trajectory 7, frame 475
trajectory 3, frame 233
trajectory 1, frame 308
trajectory 6, frame 53
trajectory 5, frame 132

Layers














Timeline

frame: 1 policy: next action:
no-op
A
B
fps
advantage
0.127
value function
9.63

Attribution

Observation Positive attribution Negative attribution

policy logits:

sums of policy logits:

Attribution legend

Click to expand feature

1
2
3
4
5
6
7
8

Hotkeys

go backwards
go forwards
toggle play/pause

Select a feature

Feature visualization

zoom in zoom out
fewer patches more patches