Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: stable-baselines3
|
| 3 |
+
tags:
|
| 4 |
+
- reinforcement-learning
|
| 5 |
+
- finrl
|
| 6 |
+
- ppo
|
| 7 |
+
- stock-trading
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
# FinRL PPO Agent (Quick Demo, 2000 steps)
|
| 11 |
+
|
| 12 |
+
Trained on DOW 30 stocks (2014-2025) using FinRL + Stable-Baselines3 PPO.
|
| 13 |
+
|
| 14 |
+
⚠️ **Toy model — only 2000 timesteps**, used to validate training pipeline.
|
| 15 |
+
Not for real trading.
|
| 16 |
+
|
| 17 |
+
## Usage
|
| 18 |
+
|
| 19 |
+
```python
|
| 20 |
+
from huggingface_hub import hf_hub_download
|
| 21 |
+
from stable_baselines3 import PPO
|
| 22 |
+
|
| 23 |
+
path = hf_hub_download(
|
| 24 |
+
repo_id="2045max/finrl-ppo-dow30-quick",
|
| 25 |
+
filename="agent_ppo.zip"
|
| 26 |
+
)
|
| 27 |
+
model = PPO.load(path)
|
| 28 |
+
```
|
| 29 |
+
|
| 30 |
+
## Training Setup
|
| 31 |
+
- Algorithm: PPO (Proximal Policy Optimization)
|
| 32 |
+
- Total timesteps: 2,000
|
| 33 |
+
- State space: 301 (cash + 30 prices + 30 holdings + 30×8 indicators)
|
| 34 |
+
- Action space: 30 (continuous, [-1, 1] per stock)
|
| 35 |
+
- Reward: portfolio value change × 1e-4
|
| 36 |
+
|
| 37 |
+
## Source
|
| 38 |
+
https://github.com/AI4Finance-Foundation/FinRL
|