# CBPOA — 武汉儿童呼吸疾病风险评估系统

FastAPI + React + `@geoscene/core` + PyTorch GCN pipeline. 预测空气质量对儿童健康的空间风险。

## Development

```bash
# Frontend (pnpm)
cd frontend && pnpm dev          # localhost:3000 → proxies /api to :8000

# Backend (Python venv)
cd backend && uvicorn main:app --reload   # localhost:8000

# ML pipeline
cd scripts && python train_model.py       # PyTorch + MLflow
```

## Where to Look

| Task | Location |
|------|----------|
| API endpoint | `backend/routers/` |
| Database / PostGIS | `backend/database.py` |
| UI component | `frontend/src/components/` |
| GeoScene map helpers | `frontend/src/geoscene/` |
| Page view | `frontend/src/pages/` |
| API client / cache | `frontend/src/services/api.ts` |
| State management | `frontend/src/stores/` |
| TypeScript types | `frontend/src/types/` |
| ETL / data processing | `scripts/` |
| ML model architecture | `models/spatiotemporal_gcn/` |
| Trained weights | `models/spatiotemporal_gcn/best_model.pt` |
| Processed features | `processed/` |
| Raw data sources | `Datas/` |
| Docker / deploy | `deploy/` |

## Data Sources

| Data | Path | Notes |
|------|------|-------|
| 气象+空气 | `Datas/气象+空气/站点_*.csv` | 3yr, 2192 files, ~2.37M rows |
| 门诊 | `Datas/view_门诊.xlsx` | 107,579 rows |
| 住院 | `Datas/view_住院.xlsx` | 5,822 rows |
| DEM高程 | `Datas/DEM/CJJJD_DEM.TIF` | 3.1GB raster |
| 人口密度 | `Datas/landscan-hd-china-v1-assets/*.tif` | 284MB |
| 行政边界 | `Datas/武汉市.geojson` | Wuhan boundary |

## ML Pipeline

```
气象(时间序列) + 站点坐标 + DEM高程 + 人口密度 → SpatialTemporalGCN → 风险预测 [1d, 3d, 7d]
```

## Agent Workflow

Explore finds → Librarian reads → You plan → Worker implements → Validator checks

Context-specific guidance lives in nested CLAUDE.md files — they load automatically when you work in those directories. Closest CLAUDE.md to the file being edited takes precedence.