Commit
·
96c14ad
1
Parent(s):
5d220d7
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,30 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
| 4 |
+
# Dataset Card for Pong-v4-expert-MCTS
|
| 5 |
+
## Table of Contents
|
| 6 |
+
- [Dataset Description](#dataset-description)
|
| 7 |
+
- [Dataset Structure](#dataset-structure)
|
| 8 |
+
- [Data Instances](#data-instances)
|
| 9 |
+
- [Data Fields](#data-fields)
|
| 10 |
+
- [Data Splits](#data-splits)
|
| 11 |
+
## Dataset Description
|
| 12 |
+
This dataset includes 8 episodes of pong-v4 environment. The expert policy is EfficientZero, which is able to generate MCTS hidden states.
|
| 13 |
+
|
| 14 |
+
## Dataset Structure
|
| 15 |
+
### Data Instances
|
| 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.
|