devdharpatel
commited on
Commit
•
f100dad
1
Parent(s):
e2a2843
Update README
Browse files
README.md
CHANGED
@@ -1,3 +1,88 @@
|
|
1 |
-
---
|
2 |
-
license: bsd-3-clause
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: bsd-3-clause
|
3 |
+
tags:
|
4 |
+
- InvertedDoublePendulum-v2
|
5 |
+
- reinforcement-learning
|
6 |
+
- decisions
|
7 |
+
- TLA
|
8 |
+
- deep-reinforcement-learning
|
9 |
+
model-index:
|
10 |
+
- name: TLA
|
11 |
+
results:
|
12 |
+
- metrics:
|
13 |
+
- type: mean_reward
|
14 |
+
value: 9356.67
|
15 |
+
name: mean_reward
|
16 |
+
- type: Action Repetition
|
17 |
+
value: .7522
|
18 |
+
name: Action Repetition
|
19 |
+
- type: Average Decisions
|
20 |
+
value: 247.76
|
21 |
+
name: Average Decisions
|
22 |
+
task:
|
23 |
+
type: OpenAI Gym
|
24 |
+
name: OpenAI Gym
|
25 |
+
dataset:
|
26 |
+
name: InvertedDoublePendulum-v2
|
27 |
+
type: InvertedDoublePendulum-v2
|
28 |
+
Paper: https://arxiv.org/abs/2305.18701
|
29 |
+
Code: https://github.com/dee0512/Temporally-Layered-Architecture
|
30 |
+
---
|
31 |
+
# Temporally Layered Architecture: InvertedDoublePendulum-v2
|
32 |
+
|
33 |
+
These are 10 trained models over **seeds (0-9)** of **[Temporally Layered Architecture (TLA)](https://github.com/dee0512/Temporally-Layered-Architecture)** agent playing **InvertedDoublePendulum-v2**.
|
34 |
+
|
35 |
+
## Model Sources
|
36 |
+
|
37 |
+
**Repository:** [https://github.com/dee0512/Temporally-Layered-Architecture](https://github.com/dee0512/Temporally-Layered-Architecture)
|
38 |
+
**Paper:** [https://doi.org/10.1162/neco_a_01718](https://doi.org/10.1162/neco_a_01718)
|
39 |
+
**Arxiv:** [arxiv.org/abs/2305.18701](https://arxiv.org/abs/2305.18701)
|
40 |
+
|
41 |
+
# Training Details:
|
42 |
+
Using the repository:
|
43 |
+
|
44 |
+
```
|
45 |
+
python main.py --env_name <environment> --seed <seed>
|
46 |
+
```
|
47 |
+
|
48 |
+
# Evaluation:
|
49 |
+
|
50 |
+
Download the models folder and place it in the same directory as the cloned repository.
|
51 |
+
Using the repository:
|
52 |
+
|
53 |
+
```
|
54 |
+
python eval.py --env_name <environment>
|
55 |
+
```
|
56 |
+
|
57 |
+
## Metrics:
|
58 |
+
|
59 |
+
**mean_reward:** Mean reward over 10 seeds
|
60 |
+
**action_repeititon:** percentage of actions that are equal to the previous action
|
61 |
+
**mean_decisions:** Number of decisions required (neural network/model forward pass)
|
62 |
+
|
63 |
+
|
64 |
+
# Citation
|
65 |
+
|
66 |
+
The paper can be cited with the following bibtex entry:
|
67 |
+
|
68 |
+
## BibTeX:
|
69 |
+
|
70 |
+
```
|
71 |
+
@article{10.1162/neco_a_01718,
|
72 |
+
author = {Patel, Devdhar and Sejnowski, Terrence and Siegelmann, Hava},
|
73 |
+
title = "{Optimizing Attention and Cognitive Control Costs Using Temporally Layered Architectures}",
|
74 |
+
journal = {Neural Computation},
|
75 |
+
pages = {1-30},
|
76 |
+
year = {2024},
|
77 |
+
month = {10},
|
78 |
+
issn = {0899-7667},
|
79 |
+
doi = {10.1162/neco_a_01718},
|
80 |
+
url = {https://doi.org/10.1162/neco\_a\_01718},
|
81 |
+
eprint = {https://direct.mit.edu/neco/article-pdf/doi/10.1162/neco\_a\_01718/2474695/neco\_a\_01718.pdf},
|
82 |
+
}
|
83 |
+
```
|
84 |
+
|
85 |
+
## APA:
|
86 |
+
```
|
87 |
+
Patel, D., Sejnowski, T., & Siegelmann, H. (2024). Optimizing Attention and Cognitive Control Costs Using Temporally Layered Architectures. Neural Computation, 1-30.
|
88 |
+
```
|