|
12 | 12 | <a href="https://claude.ai"><img src="https://img.shields.io/badge/Claude_Code-compatible-blueviolet?style=flat-square" alt="Claude Code"></a> |
13 | 13 | <img src="https://img.shields.io/badge/genes-17-gold?style=flat-square" alt="17 Genes"> |
14 | 14 | <img src="https://img.shields.io/badge/attacks-6_types-red?style=flat-square" alt="6 Attack Types"> |
15 | | - <img src="https://img.shields.io/badge/tests-22_passing-brightgreen?style=flat-square" alt="22 Tests"> |
| 15 | + <img src="https://img.shields.io/badge/tests-30_passing-brightgreen?style=flat-square" alt="30 Tests"> |
16 | 16 | <a href="https://colab.research.google.com/github/0xSanei/darwinia/blob/main/notebooks/quickstart.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open in Colab"></a> |
17 | 17 | </p> |
18 | 18 |
|
@@ -227,6 +227,8 @@ python -m darwinia evolve -d my_data.csv # Custom data |
227 | 227 | python -m darwinia evolve --json # JSON output for agents |
228 | 228 | python -m darwinia arena -c output/champions/champion_gen_0049.json |
229 | 229 | python -m darwinia arena -r 10 --json # 10 rounds, JSON output |
| 230 | +python -m darwinia validate -w 3 -g 20 # Walk-forward overfitting check |
| 231 | +python -m darwinia explain -c champion.json # Gene ablation analysis |
230 | 232 | python -m darwinia dashboard # Web UI |
231 | 233 | python -m darwinia info --json # System info as JSON |
232 | 234 | ``` |
@@ -263,6 +265,20 @@ dashboard/ # Streamlit visualization (4 pages) |
263 | 265 | scripts/ # Competitor monitoring, utilities |
264 | 266 | ``` |
265 | 267 |
|
| 268 | +### Validation & Explainability |
| 269 | + |
| 270 | +**Walk-Forward Validation** splits data chronologically (train on past, test on future). If fitness degrades out-of-sample, the strategy is overfitting. |
| 271 | + |
| 272 | +```bash |
| 273 | +python -m darwinia validate -w 3 --json |
| 274 | +``` |
| 275 | + |
| 276 | +**Gene Ablation** zeroes each gene one at a time and measures fitness drop. Genes that cause the biggest drop are the most important — proving the agent found real signal, not noise. |
| 277 | + |
| 278 | +```bash |
| 279 | +python -m darwinia explain -c champion.json --json |
| 280 | +``` |
| 281 | + |
266 | 282 | ### Three Layers |
267 | 283 |
|
268 | 284 | | Layer | Status | What It Does | |
|
0 commit comments