File size: 9,019 Bytes
49f3de8 | 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 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 | """V3 selection-era evaluation (origins 1990-2004 only).
Usage: .venv/Scripts/python.exe code/run_v3_eval.py [phase0|phase1|phase2|phase3|phase5|all]
Confirmation-era (2005-2011) numbers are produced ONLY by run_v3_confirm.py.
Every candidate in candidates_v3.yaml is run and reported, pass or fail.
"""
from __future__ import annotations
import json
import sys
import urllib.request
from pathlib import Path
import numpy as np
import pandas as pd
sys.path.insert(0, str(Path(__file__).parent))
from candidates import level_trend
from candidates_v3 import (DailyCandidates, SeasonalPrecip, ar_anomaly_temp,
build_filled_panel, damped_temp, ew_slope_forecast,
global_beta_temp, hot_delta, oni_year1_temp,
quantile_temp, volcano_clean)
from harness_v3 import SELECTION_ORIGINS, evaluate_v3, gate, skill_table_v3
from leakage import assert_no_forbidden_artifacts
from panel import Panel
from run_v2_confirm import build_packages
ROOT = Path(__file__).resolve().parent.parent
OUT = ROOT / "outputs" / "v3"
def fetch_oni() -> pd.Series | None:
"""ERSST ONI index (NOAA PSL), cached; annual means. None if offline."""
cache = ROOT / "data" / "external" / "oni.data.txt"
try:
if not cache.exists():
cache.parent.mkdir(parents=True, exist_ok=True)
with urllib.request.urlopen(
"https://psl.noaa.gov/data/correlation/oni.data", timeout=20) as r:
cache.write_bytes(r.read())
rows = {}
for line in cache.read_text(errors="replace").splitlines():
parts = line.split()
if len(parts) == 13 and parts[0].isdigit() and 1850 <= int(parts[0]) <= 2018:
vals = [float(v) for v in parts[1:]]
vals = [v for v in vals if v > -90]
if len(vals) >= 10:
rows[int(parts[0])] = float(np.mean(vals))
return pd.Series(rows).sort_index() if rows else None
except Exception as exc: # offline etc: candidate is skipped, not faked
print(f"ONI unavailable ({exc}); temp_oni_year1_shr0.5 will be skipped")
return None
class Context:
def __init__(self):
self.panel, self.packages = build_packages()
self.v2 = {m: c["v2"] for m, c in self.packages.items()}
self._daily = None
self._seasonal = None
self._filled = None
self._oni = ...
@property
def daily(self) -> DailyCandidates:
if self._daily is None:
self._daily = DailyCandidates(ROOT, self.panel)
return self._daily
@property
def seasonal(self) -> SeasonalPrecip:
if self._seasonal is None:
self._seasonal = SeasonalPrecip(ROOT, self.panel)
return self._seasonal
@property
def filled(self) -> Panel:
if self._filled is None:
self._filled = build_filled_panel(ROOT, self.panel)
return self._filled
@property
def oni(self):
if self._oni is ...:
self._oni = fetch_oni()
return self._oni
def phase_candidates(ctx: Context, phase: str) -> dict:
"""{metric: {candidate_name: fn}} including the v2 control per metric."""
p = ctx.panel
if phase == "phase0":
return {m: {"v2": ctx.v2[m]} for m in ctx.v2}
if phase == "phase1":
return {"annual_avg_temp": {
"v2": ctx.v2["annual_avg_temp"],
"damped_phi0.70_L15natW40": lambda pp, o: damped_temp(pp, o, 0.70),
"damped_phi0.85_L15natW40": lambda pp, o: damped_temp(pp, o, 0.85),
"v2temp_plus_AR_rho0.30": lambda pp, o: ar_anomaly_temp(pp, o, ctx.v2["annual_avg_temp"]),
"quantile_L15med_regW40": quantile_temp,
}}
if phase == "phase2":
d = ctx.daily
return {
"cdd": {
"v2": ctx.v2["cdd"],
"delta_bootmedian_h30_nat": lambda pp, o: d.boot_median_cdd(o),
"delta_asym1.125_h30_nat": lambda pp, o: d.delta.forecast(o, 30, "median", "nat", slope_mult=1.125),
"delta_asym1.25_h30_nat": lambda pp, o: d.delta.forecast(o, 30, "median", "nat", slope_mult=1.25),
"evanguo_L10_sens_natW40": lambda pp, o: d.evanguo_cdd(o),
},
"hottest_day_temp": {
"v2": ctx.v2["hottest_day_temp"],
"hotdelta_h30_nat": lambda pp, o: hot_delta(pp, o, 30, "nat"),
"hotdelta_h15_nat": lambda pp, o: hot_delta(pp, o, 15, "nat"),
"hotdelta_h30_reg": lambda pp, o: hot_delta(pp, o, 30, "reg"),
"blend50_hotdelta_h30_nat_mean30": lambda pp, o: 0.5 * hot_delta(pp, o, 30, "nat")
+ 0.5 * ctx.v2["hottest_day_temp"](pp, o),
},
"total_precip_in": {
"v2": ctx.v2["total_precip_in"],
"precip_seasonal_med30": lambda pp, o: ctx.seasonal.forecast(pp, o),
},
}
if phase == "phase3":
out = {
"annual_avg_temp": {
"v2": ctx.v2["annual_avg_temp"],
"temp_volcano_clean_natW40": lambda pp, o: volcano_clean(pp, o, "annual_avg_temp"),
"temp_globalbeta_shr0.5": global_beta_temp,
},
"cdd": {
"v2": ctx.v2["cdd"],
"cdd_volcano_clean_natW40": lambda pp, o: volcano_clean(pp, o, "cdd"),
"cdd_fill30_mean30": lambda pp, o: level_trend(ctx.filled, "cdd", o, 30, "mean", "none", clip_zero=True),
},
"hottest_day_temp": {
"v2": ctx.v2["hottest_day_temp"],
"hottest_fill30_mean30": lambda pp, o: level_trend(ctx.filled, "hottest_day_temp", o, 30, "mean", "none"),
},
}
if ctx.oni is not None:
out["annual_avg_temp"]["temp_oni_year1_shr0.5"] = \
lambda pp, o: oni_year1_temp(pp, o, ctx.oni, ctx.v2["annual_avg_temp"])
else:
print("skipped: temp_oni_year1_shr0.5 (no ONI data)")
return out
if phase == "phase5":
return {
"annual_avg_temp": {
"v2": ctx.v2["annual_avg_temp"],
"temp_ewslope_hl20_shr": lambda pp, o: ew_slope_forecast(pp, o, "annual_avg_temp"),
},
"cdd": {
"v2": ctx.v2["cdd"],
"cdd_ewslope_hl20_shr": lambda pp, o: ew_slope_forecast(pp, o, "cdd"),
},
}
raise ValueError(phase)
def run_phase(ctx: Context, phase: str, results: dict) -> None:
OUT.mkdir(parents=True, exist_ok=True)
registry = phase_candidates(ctx, phase)
for metric, cands in registry.items():
# Sanity probe: right shape, and finite on scored rows (the harness
# contract; unscored sparse cities may legitimately be NaN pre-2000).
origin0 = SELECTION_ORIGINS[0]
_, scored = ctx.panel.truth(metric, origin0)
for n, f in cands.items():
fc = f(ctx.panel, origin0)
assert fc.shape == (len(ctx.panel.cities), 7), f"{n} bad shape {fc.shape}"
assert np.isfinite(fc[scored]).all(), f"{n} non-finite on scored rows at {origin0}"
detail = evaluate_v3(ctx.panel, metric, cands, origins=SELECTION_ORIGINS)
detail.to_csv(OUT / f"detail_{phase}_{metric}.csv", index=False)
tab = skill_table_v3(detail)
print(f"\n== {phase} {metric} (selection 1990-2004) ==")
print(tab[tab["era"] == "selection"].round(4).to_string(index=False))
for name in cands:
if name == "v2":
results.setdefault(metric, {})["v2_selection"] = \
float(tab.loc[tab["candidate"] == "v2", "skill"].iloc[0])
continue
g = gate(detail, name, "v2", metric)
results.setdefault(metric, {})[name] = g
print(f"gate {name}: diff {g['diff']:+.4f} vs SE {g['boot_se']:.4f} "
f"-> {'PASS' if g['passes'] else 'fail'}")
def main() -> None:
assert_no_forbidden_artifacts(Path(__file__).parent)
phase = sys.argv[1] if len(sys.argv) > 1 else "all"
phases = ["phase0", "phase1", "phase2", "phase3", "phase5"] if phase == "all" else [phase]
ctx = Context()
results_path = OUT / "selection_results.json"
results = json.loads(results_path.read_text()) if results_path.exists() else {}
for ph in phases:
run_phase(ctx, ph, results)
OUT.mkdir(parents=True, exist_ok=True)
results_path.write_text(json.dumps(results, indent=2, default=float))
n_new = sum(1 for m in results.values() for k in m if k != "v2_selection")
print(f"\nnew configs evaluated so far: {n_new} (budget <= 40)")
print(f"wrote {results_path}")
if __name__ == "__main__":
main()
|