| """Compute daily/annual detection metrics and a 7x7 occlusion trend map.""" |
|
|
| import sys |
| from pathlib import Path |
|
|
| import numpy as np |
| import torch |
|
|
|
|
| ROOT = Path(__file__).resolve().parents[1] |
| sys.path.insert(0, str(ROOT)) |
| from model.precipdd import correlation, ensemble_predict, linear_trend, load_config, load_ensemble, write_json |
|
|
|
|
| def main(): |
| import matplotlib |
| matplotlib.use("Agg") |
| import matplotlib.pyplot as plt |
|
|
| config = load_config(ROOT / "conf/config.yaml") |
| values = np.load(ROOT / config["paths"]["predictions"]) |
| prediction, target, year = values["prediction"], values["target"], values["year"].astype(float) |
| if prediction.shape != target.shape or values["precipitation"].shape[1:] != (1, 55, 160): |
| raise ValueError("inference output violates scalar target or [N,1,55,160] input contract") |
| years = np.unique(year).astype(int) |
| annual_prediction = np.array([prediction[year == current].mean() for current in years]) |
| annual_target = np.array([target[year == current].mean() for current in years]) |
| threshold = config["evaluation"]["emergence_threshold_c"] |
| em_fraction = np.array([(prediction[year == current] > threshold).mean() for current in years]) |
| metrics = { |
| "daily": {"correlation": correlation(target, prediction), "rmse_c": float(np.sqrt(np.mean((prediction - target) ** 2)))}, |
| "annual": {"correlation": correlation(annual_target, annual_prediction), "rmse_c": float(np.sqrt(np.mean((annual_prediction - annual_target) ** 2)))}, |
| "emergence": {"threshold_c": threshold, "fraction_all_days": float((prediction > threshold).mean()), |
| "fraction_trend_per_decade": linear_trend(em_fraction, years.astype(float))}, |
| "trend_c_per_decade": {"daily_prediction": linear_trend(prediction, year), "annual_prediction": linear_trend(annual_prediction, years.astype(float)), |
| "annual_target": linear_trend(annual_target, years.astype(float))}, |
| "samples": {"daily": len(prediction), "years": len(years)} |
| } |
|
|
| device = torch.device("cuda" if torch.cuda.is_available() and config["runtime"]["device"] != "cpu" else "cpu") |
| models, _ = load_ensemble(ROOT / config["paths"]["checkpoint"], device) |
| count = min(config["evaluation"]["occlusion_max_days"], len(prediction)) |
| indices = np.linspace(0, len(prediction) - 1, count, dtype=int) |
| selected = torch.from_numpy(values["precipitation"][indices]).float().to(device) |
| baseline = ensemble_predict(models, selected).cpu().numpy() |
| selected_year = year[indices] |
| patch, stride = config["evaluation"]["occlusion_patch"], config["evaluation"]["occlusion_stride"] |
| half = patch // 2 |
| sensitivity = np.zeros((55, 160), dtype=np.float32) |
| for lat_start in range(0, 55, stride): |
| for lon_start in range(0, 160, stride): |
| masked = selected.clone() |
| lat_stop, lon_stop = min(lat_start + patch, 55), min(lon_start + patch, 160) |
| masked[:, :, lat_start:lat_stop, lon_start:lon_stop] = 0.0 |
| delta = baseline - ensemble_predict(models, masked).cpu().numpy() |
| score = linear_trend(delta, selected_year) if np.ptp(selected_year) else float(delta.mean()) |
| sensitivity[lat_start:min(lat_start + stride, 55), lon_start:min(lon_start + stride, 160)] = score |
| metrics["occlusion"] = {"patch": [patch, patch], "map_shape": [55, 160], "stride": stride, "sampled_days": count, |
| "quantity": "AGMT occlusion-sensitivity trend in degC per decade"} |
| write_json(ROOT / config["paths"]["evaluation_metrics"], metrics) |
|
|
| fig, axes = plt.subplots(3, 1, figsize=(11, 11), constrained_layout=True) |
| axes[0].plot(years, annual_target, color="#202020", label="target AGMT") |
| axes[0].plot(years, annual_prediction, color="#c84c32", label="DD estimate") |
| axes[0].axhline(threshold, color="#777777", linestyle="--", label="0.42 C EM threshold") |
| axes[0].set(ylabel="AGMT anomaly (C)", title="Annual mean of daily estimates") |
| axes[0].legend(ncol=3) |
| axes[1].plot(years, em_fraction, color="#196f82") |
| axes[1].set(xlabel="Year", ylabel="Fraction", ylim=(-0.03, 1.03), title="Emergence days (estimated AGMT > 0.42 C)") |
| limit = float(np.max(np.abs(sensitivity))) or 1e-6 |
| image = axes[2].imshow(sensitivity, origin="lower", aspect="auto", extent=(0, 400, values["latitude"][0], values["latitude"][-1]), |
| cmap="RdBu_r", vmin=-limit, vmax=limit) |
| axes[2].set(xlabel="Longitude (degrees E, extended)", ylabel="Latitude", title="7x7 occlusion-sensitivity trend") |
| fig.colorbar(image, ax=axes[2], label="C decade-1") |
| figure = ROOT / config["paths"]["comparison_figure"] |
| figure.parent.mkdir(parents=True, exist_ok=True) |
| fig.savefig(figure, dpi=160) |
| plt.close(fig) |
| print(f"metrics={config['paths']['evaluation_metrics']} figure={config['paths']['comparison_figure']}") |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
