refactor: extract trend_engine, unify probability engine across bot and web, add unit tests
This commit is contained in:
2569718930@qq.com
+29
-135
@@ -268,126 +268,42 @@ def _analyze(city: str) -> Dict[str, Any]:
|
||||
else:
|
||||
peak_status = "before"
|
||||
|
||||
# ── 10. Probability distribution ──
|
||||
# ── 10. Shared analysis (probability, trend, AI) via trend_engine ──
|
||||
# This single call replaces the duplicate probability engine, dead market
|
||||
# detection, forecast bust grading, and AI context building.
|
||||
from src.analysis.trend_engine import analyze_weather_trend as _trend_analyze
|
||||
from src.analysis.ai_analyzer import get_ai_analysis
|
||||
|
||||
probabilities = []
|
||||
mu = None
|
||||
forecast_miss_deg = 0 # How far actual is below forecasts
|
||||
if (
|
||||
ens_data["p10"] is not None
|
||||
and ens_data["p90"] is not None
|
||||
and ens_data["median"] is not None
|
||||
):
|
||||
sigma = (ens_data["p90"] - ens_data["p10"]) / 2.56
|
||||
if sigma < 0.1:
|
||||
sigma = 0.1
|
||||
ai_text = ""
|
||||
try:
|
||||
_, ai_context, sd = _trend_analyze(raw, sym, city)
|
||||
|
||||
# Historical MAE floor
|
||||
acc = get_deb_accuracy(city)
|
||||
if acc:
|
||||
_, hist_mae, _, _ = acc
|
||||
if hist_mae > sigma:
|
||||
sigma = hist_mae
|
||||
# Use structured data from shared engine
|
||||
mu = sd.get("mu")
|
||||
probabilities = sd.get("probabilities", [])
|
||||
trend_info["is_dead_market"] = sd.get("trend_info", {}).get("is_dead_market", False)
|
||||
trend_info["direction"] = sd.get("trend_info", {}).get("direction", trend_info.get("direction", "unknown"))
|
||||
trend_info["is_cooling"] = sd.get("trend_info", {}).get("is_cooling", False)
|
||||
peak_status = sd.get("peak_status", peak_status)
|
||||
|
||||
# Shock score
|
||||
recent_obs = metar.get("recent_obs", []) if metar else []
|
||||
shock = 0.0
|
||||
if len(recent_obs) >= 2:
|
||||
o_obs, n_obs = recent_obs[-1], recent_obs[0]
|
||||
wd_o, wd_n = _sf(o_obs.get("wdir")), _sf(n_obs.get("wdir"))
|
||||
ws_n = _sf(n_obs.get("wspd")) or 0
|
||||
if wd_o is not None and wd_n is not None:
|
||||
ad = abs(wd_n - wd_o)
|
||||
if ad > 180:
|
||||
ad = 360 - ad
|
||||
shock += min(ad / 90, 1) * min(ws_n / 15, 1) * 0.4
|
||||
cr_o = o_obs.get("cloud_rank", 0)
|
||||
cr_n = n_obs.get("cloud_rank", 0)
|
||||
shock += min(abs(cr_n - cr_o) / 3, 1) * 0.35
|
||||
ap_o, ap_n = _sf(o_obs.get("altim")), _sf(n_obs.get("altim"))
|
||||
if ap_o is not None and ap_n is not None:
|
||||
shock += min(abs(ap_n - ap_o) / 4, 1) * 0.25
|
||||
if shock > 0.05:
|
||||
sigma *= 1 + 0.5 * shock
|
||||
# Use shared DEB if not already set
|
||||
if deb_val is None and sd.get("deb_prediction") is not None:
|
||||
deb_val = sd["deb_prediction"]
|
||||
deb_weights = sd.get("deb_weights", "")
|
||||
|
||||
# Time-based sigma adjustment
|
||||
if local_hour_frac > last_peak_h:
|
||||
sigma *= 0.3
|
||||
elif first_peak_h <= local_hour_frac <= last_peak_h:
|
||||
sigma *= 0.7
|
||||
|
||||
# Mu calculation — reality-anchored
|
||||
forecast_highs = [h for h in current_forecasts.values() if h is not None]
|
||||
forecast_median = (
|
||||
sorted(forecast_highs)[len(forecast_highs) // 2]
|
||||
if forecast_highs
|
||||
else ens_data["median"]
|
||||
)
|
||||
|
||||
# Compute forecast miss magnitude
|
||||
if max_so_far is not None and forecast_median is not None:
|
||||
forecast_miss_deg = round(forecast_median - max_so_far, 1)
|
||||
|
||||
# --- Key fix: Reality-anchored μ ---
|
||||
# If we are past or in the peak window AND actual max is significantly
|
||||
# below forecasts, anchor μ on max_so_far, not on forecast_median.
|
||||
if (
|
||||
max_so_far is not None
|
||||
and forecast_median is not None
|
||||
and peak_status in ("past", "in_window")
|
||||
and max_so_far < forecast_median - 2.0
|
||||
):
|
||||
# Forecast bust: μ anchors on actual max, not failed predictions
|
||||
# Allow small upward margin only if still warming
|
||||
if trend_info["is_cooling"] or peak_status == "past":
|
||||
mu = max_so_far
|
||||
else:
|
||||
# Still in window and warming — small margin
|
||||
mu = max_so_far + 0.5
|
||||
else:
|
||||
# Normal case: blend forecast and ensemble
|
||||
mu = (
|
||||
forecast_median * 0.7 + ens_data["median"] * 0.3
|
||||
if forecast_median is not None
|
||||
else ens_data["median"]
|
||||
# Append multi-model divergence for AI
|
||||
if current_forecasts and ai_context:
|
||||
mm_str = " | ".join(
|
||||
[f"{k}:{v}{sym}" for k, v in current_forecasts.items() if v]
|
||||
)
|
||||
if max_so_far is not None and max_so_far > mu:
|
||||
mu = max_so_far + (0.3 if not trend_info["is_cooling"] else 0.0)
|
||||
ai_context += f"\n模型分歧: {mm_str}"
|
||||
|
||||
def _norm_cdf(x, m, s):
|
||||
return 0.5 * (1 + math.erf((x - m) / (s * math.sqrt(2))))
|
||||
|
||||
min_wu = round(max_so_far) if max_so_far is not None else -999
|
||||
probs = {}
|
||||
for n in range(round(mu) - 2, round(mu) + 3):
|
||||
if n < min_wu:
|
||||
continue
|
||||
p = _norm_cdf(n + 0.5, mu, sigma) - _norm_cdf(n - 0.5, mu, sigma)
|
||||
if p > 0.01:
|
||||
probs[n] = p
|
||||
total = sum(probs.values())
|
||||
if total > 0:
|
||||
probs = {k: v / total for k, v in probs.items()}
|
||||
for t, p in sorted(probs.items(), key=lambda x: x[1], reverse=True)[:4]:
|
||||
probabilities.append(
|
||||
{"value": t, "range": f"[{t-0.5}~{t+0.5})", "probability": round(p, 3)}
|
||||
)
|
||||
|
||||
# ── 11. Dead market detection ──
|
||||
is_dead = False
|
||||
if max_so_far is not None and cur_temp is not None:
|
||||
if local_hour >= 21 and max_so_far - cur_temp >= 3.0:
|
||||
is_dead = True
|
||||
elif local_hour > last_peak_h and max_so_far - cur_temp >= 1.5:
|
||||
is_dead = True
|
||||
trend_info["is_dead_market"] = is_dead
|
||||
|
||||
# Override probabilities when dead market confirmed
|
||||
if is_dead and max_so_far is not None:
|
||||
settled = round(max_so_far)
|
||||
mu = max_so_far
|
||||
probabilities = [
|
||||
{"value": settled, "range": f"[{settled-0.5}~{settled+0.5})", "probability": 1.0}
|
||||
]
|
||||
if ai_context:
|
||||
ai_text = get_ai_analysis(ai_context, city, sym)
|
||||
except Exception as e:
|
||||
logger.warning(f"Analysis/AI skipped for {city}: {e}")
|
||||
|
||||
# ── 12. Hourly data (today only, for chart) ──
|
||||
today_hourly: Dict[str, list] = {"times": [], "temps": [], "radiation": []}
|
||||
@@ -442,28 +358,6 @@ def _analyze(city: str) -> Dict[str, Any]:
|
||||
"rain_24h": _sf(mgc.get("rain_24h")),
|
||||
}
|
||||
|
||||
# ── 15. AI Analysis (reuse bot_listener's analyze_weather_trend) ──
|
||||
ai_text = ""
|
||||
try:
|
||||
from src.analysis.ai_analyzer import get_ai_analysis
|
||||
from bot_listener import analyze_weather_trend
|
||||
|
||||
# analyze_weather_trend returns (display_str, ai_context)
|
||||
# It already handles: DEB, probability, dead market, forecast bust,
|
||||
# METAR trend, peak window, wind, cloud, humidity, etc.
|
||||
_, ai_context = analyze_weather_trend(raw, sym, city)
|
||||
|
||||
# Append multi-model divergence if not already included
|
||||
if current_forecasts and ai_context:
|
||||
mm_str = " | ".join(
|
||||
[f"{k}:{v}{sym}" for k, v in current_forecasts.items() if v]
|
||||
)
|
||||
ai_context += f"\n模型分歧: {mm_str}"
|
||||
|
||||
if ai_context:
|
||||
ai_text = get_ai_analysis(ai_context, city, sym)
|
||||
except Exception as e:
|
||||
logger.warning(f"AI analysis skipped for {city}: {e}")
|
||||
|
||||
# ── Assemble result ──
|
||||
result = {
|
||||
|
||||
Reference in New Issue
Block a user