Datasets:

ArXiv:
License:
kxzxvbk commited on
Commit
e422806
1 Parent(s): b61ce65

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +6 -6
README.md CHANGED
@@ -16,15 +16,15 @@ This dataset includes 8 episodes of pong-v4 environment. The expert policy is Ef
16
  A data point comprises tuples of sequences of (observations, actions, hidden_states):
17
  ```
18
  {
19
- "obs":datasets.Array2D(),
20
- "actions":datasets.Array2D(),
21
- "hidden_state":datasets.Array2D(),
22
  }
23
  ```
24
  ### Data Fields
25
- - `obs`: An Array2D containing observations from 8 trajectories of an evaluated agent.
26
- - `actions`: An Array2D containing actions from 8 trajectories of an evaluated agent.
27
- - `hidden_state`: An Array2D containing corresponding hidden states generated by EfficientZero, from 8 trajectories of an evaluated agent.
28
 
29
  ### Data Splits
30
  There is only a training set for this dataset, as evaluation is undertaken by interacting with a simulator.
 
16
  A data point comprises tuples of sequences of (observations, actions, hidden_states):
17
  ```
18
  {
19
+ "obs":datasets.Array3D(),
20
+ "actions":int,
21
+ "hidden_state":datasets.Array3D(),
22
  }
23
  ```
24
  ### Data Fields
25
+ - `obs`: An Array3D containing observations from 8 trajectories of an evaluated agent. The data type is uint8 and each value is in 0 to 255.
26
+ - `actions`: An integer containing actions from 8 trajectories of an evaluated agent. This value is from 0 to 5.
27
+ - `hidden_state`: An Array3D containing corresponding hidden states generated by EfficientZero, from 8 trajectories of an evaluated agent. The data type is float32.
28
 
29
  ### Data Splits
30
  There is only a training set for this dataset, as evaluation is undertaken by interacting with a simulator.