marvinandClaude Opus 5.5
9ffee8b952
- New docs/09-limits-and-performance.md: principles, all protection layers, what players see, configuration reference, measurements, tuning guide, monitoring and how to test - Server chapter summarises the limits and links to chapter 9; circuit breaker description and AI_MAX_INFLIGHT default corrected - Architecture: new files in the layout, design decisions for load handling; docs index lists chapter 9 - Protocol, AI and operations chapters link to chapter 9; test counts - Root and component READMEs describe the limits and settings Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
32 lines
1.8 KiB
Markdown
32 lines
1.8 KiB
Markdown
# wizard_ai
|
||
|
||
AI player for the Wizard server. The actual RL model is still to be implemented; this package
|
||
provides everything around it.
|
||
|
||
- `agents/` — `Agent` interface (`act(observation) -> action`) plus `RandomAgent`,
|
||
`HeuristicAgent` (baseline / fallback) and `ModelAgent` (wraps a `Policy`)
|
||
- `model/encoding.py` — observation → fixed-size feature vector (`FEATURE_SIZE`), and the fixed
|
||
action space (21 bids + 60 cards) with legal-action masks; works for 3–6 players
|
||
- `model/policy.py` — **the place to plug in the model**: subclass `Policy`, implement
|
||
`select`, `update`, `save_weights`/`load`, register with `@register_policy`.
|
||
`UniformPolicy` is a placeholder so the pipeline runs end to end.
|
||
- `training/env.py` — turn-based multi-agent environment around `wizard_game`
|
||
(reward = score change per round / 100)
|
||
- `training/self_play.py` — self-play loop collecting per-player trajectories and calling
|
||
`Policy.update`, with checkpointing
|
||
- `training/evaluate.py` — play agents against each other (mean score, win rate)
|
||
- `service/client.py` — production: connects to `ws://wizard_server:8083/internal/ai`,
|
||
answers `decide` requests, reconnects automatically. Work is bounded: `AI_WORKERS` (1)
|
||
parallel decisions, at most `AI_MAX_QUEUE` (64) requests in progress (more are answered
|
||
"overloaded"), expired requests are dropped, and from `AI_DEGRADE_QUEUE` (8) waiting requests —
|
||
or when the model is too slow — the heuristic decides instead of the model
|
||
(see `../docs/09-limits-and-performance.md`)
|
||
|
||
```bash
|
||
pip install -r requirements-dev.txt
|
||
python -m wizard_ai train --episodes 5000 --out models/latest
|
||
python -m wizard_ai evaluate --agents model,heuristic,random,random --games 200
|
||
python -m wizard_ai serve # env: SERVER_URL, AI_SHARED_SECRET, AGENT, MODEL_PATH, AI_WORKERS, AI_MAX_QUEUE, AI_DEGRADE_QUEUE
|
||
pytest
|
||
```
|