Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- chess
|
| 4 |
+
- pytorch
|
| 5 |
+
- alphazero
|
| 6 |
+
- game
|
| 7 |
+
---
|
| 8 |
+
|
| 9 |
+
# Chess AlphaZero (PyTorch)
|
| 10 |
+
|
| 11 |
+
Chess policy & value network trained from scratch on 1.2M positions from Lichess.
|
| 12 |
+
|
| 13 |
+
## Architecture
|
| 14 |
+
- Residual CNN (6 blocks, 128 filters, 10M parameters)
|
| 15 |
+
- Policy head: move probability over 4096 possible moves
|
| 16 |
+
- Value head: win probability in [-1, +1]
|
| 17 |
+
- MCTS: 100 simulations per move
|
| 18 |
+
|
| 19 |
+
## Training
|
| 20 |
+
- Dataset: Lichess (20,000+ games, 1.2M positions)
|
| 21 |
+
- Epochs: 20
|
| 22 |
+
- Final validation accuracy: 42.7%
|
| 23 |
+
- Framework: PyTorch
|