"""Risk Mapper — produce a flood risk report from a trained FloodLSTM.""" from __future__ import annotations import json import sys from datetime import datetime, timezone from pathlib import Path from typing import Any import numpy as np import torch if __package__ is None or __package__ == "": sys.path.append(str(Path(__file__).resolve().parents[1])) from src.config import load_default_config, resolve_project_path from src.pipeline.saving.checkpoint import load_model def risk_class(prob: float) -> str: if prob < 0.30: return "Low" if prob <= 0.70: return "Moderate" return "High" def confidence(prob: float) -> float: return abs(prob - 0.5) * 2.0 def _make_tabular_stub(cfg: dict[str, Any], device: torch.device) -> torch.Tensor: """Return a random normalised input tensor for demo inference. Shape: (1, window_size, 20) — mimics one sliding-window sample. If `inference.default_features` points to a JSON file containing a dict of {feature_name: value} pairs, those values are used instead. """ from src.pipeline.ingestion.loader import FEATURE_COLUMNS # noqa: PLC0415 window_size = int(cfg["data"]["window_size"]) num_features = len(FEATURE_COLUMNS) feature_path_str = str(cfg.get("inference", {}).get("default_features") or "") if feature_path_str: feat_path = resolve_project_path(feature_path_str) if feat_path.exists(): raw = json.loads(feat_path.read_text(encoding="utf-8")) row = np.array( [float(raw.get(col, 0.0)) for col in FEATURE_COLUMNS], dtype=np.float32, ) # Tile the same feature vector over the window seq = np.tile(row, (window_size, 1)) return torch.from_numpy(seq).unsqueeze(0).to(device) # (1, W, F) # No features file → use random stub rng = np.random.default_rng(int(cfg["project"]["seed"])) seq = rng.standard_normal((window_size, num_features)).astype(np.float32) return torch.from_numpy(seq).unsqueeze(0).to(device) def main() -> None: cfg = load_default_config() for key in ("scaler", "checkpoints", "insight_reports"): cfg["paths"][key] = str(resolve_project_path(cfg["paths"][key])) device = torch.device("cuda" if torch.cuda.is_available() else "cpu") model = load_model(cfg, device) features = _make_tabular_stub(cfg, device) with torch.no_grad(): prob = float(model(features).item()) report = { "timestamp_utc": datetime.now(timezone.utc).isoformat(), "probability": round(prob, 6), "risk_class": risk_class(prob), "confidence": round(confidence(prob), 6), "model_version": cfg["project"]["name"], "window_size": cfg["data"]["window_size"], "flood_threshold": cfg["data"]["flood_threshold"], "data_provenance": { "dataset": "flood.csv (Kaggle GFD tabular)", "features": 20, "label": "FloodProbability (regression)", }, } infer_cfg = cfg.get("inference", {}) output_str = str(infer_cfg.get("output_path", "")).strip() if output_str: output = resolve_project_path(output_str) else: output = ( resolve_project_path(cfg["paths"]["insight_reports"]) / f"risk_report_{datetime.now(timezone.utc).strftime('%Y%m%dT%H%M%S')}.json" ) output.parent.mkdir(parents=True, exist_ok=True) output.write_text(json.dumps(report, indent=2), encoding="utf-8") print(json.dumps(report, indent=2)) if __name__ == "__main__": main()