Datasets:

ArXiv:
License:
kxzxvbk commited on
Commit
773b0e7
1 Parent(s): e422806

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +54 -6
README.md CHANGED
@@ -3,12 +3,14 @@ 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
@@ -21,10 +23,56 @@ A data point comprises tuples of sequences of (observations, actions, hidden_sta
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.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  ---
4
  # Dataset Card for Pong-v4-expert-MCTS
5
  ## Table of Contents
6
+ [TOC]
7
+
8
+ ## Supported Tasks and Leaderboard
9
+
10
+ - TODO
11
+
12
  ## Dataset Description
13
+
14
  This dataset includes 8 episodes of pong-v4 environment. The expert policy is EfficientZero, which is able to generate MCTS hidden states.
15
 
16
  ## Dataset Structure
 
23
  "hidden_state":datasets.Array3D(),
24
  }
25
  ```
26
+ ## Source Data
27
+
28
  ### Data Fields
29
+
30
  - `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.
31
  - `actions`: An integer containing actions from 8 trajectories of an evaluated agent. This value is from 0 to 5.
32
  - `hidden_state`: An Array3D containing corresponding hidden states generated by EfficientZero, from 8 trajectories of an evaluated agent. The data type is float32.
33
 
34
  ### Data Splits
35
+ There is only a training set for this dataset, as evaluation is undertaken by interacting with a simulator.
36
+
37
+ ## Data Creation
38
+
39
+ ### Curation Rationale
40
+
41
+ - This dataset includes expert data generated by EfficientZero. Since it contains hidden states for each observation, it is suitable for Imitation Learning methods that learn from a sequence like [Procedure Cloning (PC)](https://arxiv.org/abs/2205.10816).
42
+ ### Source Data
43
+ #### Initial Data Collection and Normalization
44
+ - This dataset is collected by EfficientZero policy.
45
+ - Each return of 8 episodes is 20.
46
+ - No normalization is previously applied ( i.e. each value of observation is a uint8 scalar in [0, 255] )
47
+ #### Who are the source language producers?
48
+ - [@kxzxvbk](https://huggingface.co/kxzxvbk)
49
+
50
+ #### Annotations
51
+ - The format of observation picture is [H, W, C], where the channel dimension is the last dimension of the tensor.
52
+
53
+ ## Considerations for Using the Data
54
+
55
+ ### Social Impact of Dataset
56
+
57
+ - This dataset can be used for Imitation Learning, especially for algorithms that learn from a sequence.
58
+ - Very few dataset is open-sourced currently for MCTS based policy.
59
+ - This dataset can potentially promote the research for sequence based imitation learning algorithms.
60
+
61
+ ### Known Limitations
62
+ - TODO
63
+
64
+ ## Additional Information
65
+
66
+ ### Licensing Information
67
+ - TODO
68
+
69
+ ### Citation Information
70
+
71
+ ```
72
+ TODO
73
+ ```
74
+
75
+ ### Contributions
76
+ Thanks to [@test](test_url), for adding this dataset.
77
+
78
+ [How to contribute to Datasets](https://files.pushshift.io/reddit/)