File size: 3,426 Bytes
d82626f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
"""Plausibility audit: every forecast vs the city's own observed history.



For each city-metric: historical (valid, <=2018) min/max/mean/sd and count,

then forecast exceedance over the historical record, z-score of the 2025

forecast, and the implied 2019->2025 change. Flags rows worth hand-checking.

Writes outputs/audit_forecast_plausibility.csv and prints the worst flags.

"""
from __future__ import annotations

import sys
from pathlib import Path

import numpy as np
import pandas as pd

sys.path.insert(0, str(Path(__file__).parent))
from panel import Panel

ROOT = Path(__file__).resolve().parent.parent


def main() -> None:
    panel = Panel(ROOT)
    fc = pd.read_csv(ROOT / "submission.csv", dtype={"city_id": str})
    rows = []
    for metric, vals in panel.values.items():
        with np.errstate(invalid="ignore"):
            hist_min = np.nanmin(np.where(np.isfinite(vals), vals, np.nan), axis=1)
            hist_max = np.nanmax(np.where(np.isfinite(vals), vals, np.nan), axis=1)
            hist_mean = np.nanmean(np.where(np.isfinite(vals), vals, np.nan), axis=1)
            hist_sd = np.nanstd(np.where(np.isfinite(vals), vals, np.nan), axis=1)
        n_hist = np.isfinite(vals).sum(axis=1)
        sub = fc[fc["metric"] == metric]
        piv = sub.pivot(index="city_id", columns="year", values="predicted_value") \
                 .reindex(index=panel.cities)
        f19, f25 = piv[2019].to_numpy(), piv[2025].to_numpy()
        rows.append(pd.DataFrame({
            "city_id": panel.cities, "metric": metric, "n_hist": n_hist,
            "hist_min": hist_min, "hist_max": hist_max, "hist_mean": hist_mean,
            "hist_sd": hist_sd, "fc_2019": f19, "fc_2025": f25,
            "fc2025_minus_histmax": f25 - hist_max,
            "fc2025_z": (f25 - hist_mean) / np.where(hist_sd > 0, hist_sd, np.nan),
            "fc_change_19_25": f25 - f19,
        }))
    out = pd.concat(rows, ignore_index=True)
    out.to_csv(ROOT / "outputs" / "audit_forecast_plausibility.csv", index=False)

    print("=== forecast ranges by metric (all cities/years) ===")
    print(fc.groupby("metric")["predicted_value"].describe().round(2).to_string())
    print("\n=== worst record-exceedance: fc2025 - historical max (per metric) ===")
    for metric, sub in out.groupby("metric"):
        top = sub.reindex(sub["fc2025_minus_histmax"].sort_values(ascending=False).index).head(6)
        print(f"\n-- {metric} --")
        print(top[["city_id", "n_hist", "hist_max", "hist_mean", "hist_sd",
                   "fc_2019", "fc_2025", "fc2025_minus_histmax", "fc2025_z"]]
              .round(2).to_string(index=False))
    print("\n=== largest |z| of 2025 forecast vs own history ===")
    for metric, sub in out.groupby("metric"):
        top = sub.reindex(sub["fc2025_z"].abs().sort_values(ascending=False).index).head(4)
        print(f"-- {metric}: " + "; ".join(
            f"{r.city_id} z={r.fc2025_z:+.1f} (hist {r.hist_mean:.0f}±{r.hist_sd:.0f}, n={int(r.n_hist)})"
            for r in top.itertuples()))
    print("\n=== sparsest histories (n_hist < 15) and their 2025 forecasts ===")
    sparse = out[out["n_hist"] < 15].sort_values("n_hist")
    print(sparse[["city_id", "metric", "n_hist", "hist_mean", "fc_2025"]].round(1)
          .to_string(index=False) if len(sparse) else "none")


if __name__ == "__main__":
    main()