mirror of
https://github.com/wassname/pytorch-soft-actor-critic.git
synced 2026-08-12 12:20:51 +08:00
78 lines
2.2 KiB
Markdown
78 lines
2.2 KiB
Markdown
### Description
|
||
------------
|
||
Reimplementation of [Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement
|
||
Learning with a Stochastic Actor](https://arxiv.org/pdf/1801.01290.pdf).
|
||
|
||
Contributions are welcome. If you find any mistake (very likely) or know how to make it more stable, don't hesitate to send a pull request.
|
||
|
||
### Requirements
|
||
------------
|
||
|
||
- [mujoco-py](https://github.com/openai/mujoco-py)
|
||
- [TensorboardX](https://github.com/lanpa/tensorboardX)
|
||
- [PyTorch](http://pytorch.org/)
|
||
|
||
### Run
|
||
------------
|
||
|
||
#### For SAC :
|
||
|
||
```
|
||
python main.py --env-name Humanoid-v2 --scale_R 20
|
||
```
|
||
|
||
#### For SAC (Hard Update):
|
||
|
||
```
|
||
python main.py --env-name Humanoid-v2 --scale_R 20 --tau 1 --target_update_interval 1000
|
||
```
|
||
|
||
#### For SAC (Deterministic, Hard Update):
|
||
|
||
```
|
||
python main.py --env-name Humanoid-v2 --scale_R 20 --deterministic True --tau 1 --target_update_interval 1000
|
||
```
|
||
|
||
### Results
|
||
------------
|
||
My results on Humanoid-v2 environment using SAC, SAC(hard update) and SAC(deterministic, hard update).
|
||
This is a plot of average rewards at every 10000 step interval
|
||
|
||
|
||
|
||
### Parameters
|
||
-------------
|
||
|
||
|
||
| Parameters | Value |
|
||
| --------------- | ------------- |
|
||
|**Shared**|-|
|
||
| optimizer | Adam |
|
||
| learning rate(`--lr`) | 3x10<sup>−4</sup> |
|
||
| discount(`--gamma`) (γ) | 0.99 |
|
||
| replay buffer size(`--replay_size`) | 1x10<sup>6</sup> |
|
||
|number of hidden layers (all networks)|2|
|
||
|number of hidden units per layer(`--hidden_size`)|256|
|
||
|number of samples per minibatch(`--batch_size`)|256|
|
||
|nonlinearity|ReLU|
|
||
|**SAC**|-|
|
||
|target smoothing coefficient(`--tau`) (τ)|0.005|
|
||
|target update interval(`--target_update_interval`)|1|
|
||
|gradient steps(`--updates_per_step`)|1|
|
||
|**SAC** *(Hard Update)*|-|
|
||
|target smoothing coefficient(`--tau`) (τ)|1|
|
||
|target update interval(`--target_update_interval`)|1000|
|
||
|gradient steps (except humanoids)(`--updates_per_step`)|4|
|
||
|gradient steps (humanoids)(`--updates_per_step`)|1|
|
||
|
||
|
||
|
||
|
||
| Environment **(`--env-name`)**| Reward Scale **(`--scale_R`)**|
|
||
| --------------- | ------------- |
|
||
| HalfCheetah-v2 | 5 |
|
||
| Hopper-v2 | 5 |
|
||
| Walker2d-v2 | 5 |
|
||
| Ant-v2 | 5 |
|
||
| Humanoid-v2 | 20 |
|