Datasets:

ArXiv:
License:
kxzxvbk commited on
Commit
10ca435
1 Parent(s): bf9c32b

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +0 -8
README.md CHANGED
@@ -3,7 +3,6 @@ license: apache-2.0
3
  ---
4
  # Dataset Card for Pong-v4-expert-MCTS
5
  ## Table of Contents
6
-
7
  - [Supported Tasks and Baseline](#support-tasks-and-baseline)
8
 
9
  - [Data Usage](#data-usage)
@@ -24,10 +23,8 @@ license: apache-2.0
24
  - [Contributions](##Contributions)
25
 
26
  ## Supported Tasks and Baseline
27
-
28
  - This dataset supports the training for [Procedure Cloning (PC )](https://arxiv.org/abs/2205.10816) algorithm.
29
  - Baselines when sequence length for decision is 0:
30
-
31
  | Train loss | Test Acc | Reward |
32
  | -------------------------------------------------- | -------- | ------ |
33
  | ![feature](./sup_loss.png) | 0.90 | 20 |
@@ -36,13 +33,9 @@ license: apache-2.0
36
  | ----------------------------------------------------- | ------------------------------------------------- | --------------------------------------------------- | ------ |
37
  | ![feature](./action_loss.png) | ![feature](./hs_loss.png) | ![feature](./train_acc.png) | -21 |
38
  ## Data Usage
39
-
40
  ### Data description
41
-
42
  This dataset includes 8 episodes of pong-v4 environment. The expert policy is [EfficientZero](https://arxiv.org/abs/2111.00210), which is able to generate MCTS hidden states. Because of the contained hidden states for each observation, this dataset is suitable for Imitation Learning methods that learn from a sequence like PC.
43
-
44
  ### Data Fields
45
-
46
  - `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. The shape of this tensor is [96, 96, 3], that is, the channel dimension in the last dimension.
47
  - `actions`: An integer containing actions from 8 trajectories of an evaluated agent. This value is from 0 to 5. Details about this environment can be viewed at [Pong - Gym Documentation](https://www.gymlibrary.dev/environments/atari/pong/).
48
  - `hidden_state`: An Array3D containing corresponding hidden states generated by EfficientZero, from 8 trajectories of an evaluated agent. The data type is float32.
@@ -70,7 +63,6 @@ def generate_examples(self, filepath):
70
  There is only a training set for this dataset, as evaluation is undertaken by interacting with a simulator.
71
 
72
  ### Initial Data Collection and Normalization
73
-
74
  - This dataset is collected by EfficientZero policy.
75
  - The standard for expert data is that each return of 8 episodes is over 20.
76
  - No normalization is previously applied ( i.e. each value of observation is a uint8 scalar in [0, 255] )
 
3
  ---
4
  # Dataset Card for Pong-v4-expert-MCTS
5
  ## Table of Contents
 
6
  - [Supported Tasks and Baseline](#support-tasks-and-baseline)
7
 
8
  - [Data Usage](#data-usage)
 
23
  - [Contributions](##Contributions)
24
 
25
  ## Supported Tasks and Baseline
 
26
  - This dataset supports the training for [Procedure Cloning (PC )](https://arxiv.org/abs/2205.10816) algorithm.
27
  - Baselines when sequence length for decision is 0:
 
28
  | Train loss | Test Acc | Reward |
29
  | -------------------------------------------------- | -------- | ------ |
30
  | ![feature](./sup_loss.png) | 0.90 | 20 |
 
33
  | ----------------------------------------------------- | ------------------------------------------------- | --------------------------------------------------- | ------ |
34
  | ![feature](./action_loss.png) | ![feature](./hs_loss.png) | ![feature](./train_acc.png) | -21 |
35
  ## Data Usage
 
36
  ### Data description
 
37
  This dataset includes 8 episodes of pong-v4 environment. The expert policy is [EfficientZero](https://arxiv.org/abs/2111.00210), which is able to generate MCTS hidden states. Because of the contained hidden states for each observation, this dataset is suitable for Imitation Learning methods that learn from a sequence like PC.
 
38
  ### Data Fields
 
39
  - `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. The shape of this tensor is [96, 96, 3], that is, the channel dimension in the last dimension.
40
  - `actions`: An integer containing actions from 8 trajectories of an evaluated agent. This value is from 0 to 5. Details about this environment can be viewed at [Pong - Gym Documentation](https://www.gymlibrary.dev/environments/atari/pong/).
41
  - `hidden_state`: An Array3D containing corresponding hidden states generated by EfficientZero, from 8 trajectories of an evaluated agent. The data type is float32.
 
63
  There is only a training set for this dataset, as evaluation is undertaken by interacting with a simulator.
64
 
65
  ### Initial Data Collection and Normalization
 
66
  - This dataset is collected by EfficientZero policy.
67
  - The standard for expert data is that each return of 8 episodes is over 20.
68
  - No normalization is previously applied ( i.e. each value of observation is a uint8 scalar in [0, 255] )