predictor / services /model /wcmodel.py
pengpeng9099's picture
Deploy World Cup 2026 Predictor (docker space)
1790bde verified
Raw History Blame Contribute Delete
4.47 kB
"""Dixon-Coles goal model for international football, with time decay and
neutral-venue handling. Wraps penaltyblog."""
from __future__ import annotations
import json
from dataclasses import dataclass
import numpy as np
import pandas as pd
import penaltyblog as pb
import config
from db import connect
def load_training_frame(train_from: str = config.TRAIN_FROM) -> pd.DataFrame:
"""Played matches since `train_from`, restricted to teams we can model."""
with connect() as conn:
df = pd.read_sql_query(
"SELECT date,home_team,away_team,home_score,away_score,neutral "
"FROM matches WHERE home_score IS NOT NULL AND date >= ? "
"ORDER BY date",
conn, params=(train_from,),
)
df["date"] = pd.to_datetime(df["date"])
return df
@dataclass
class WorldCupModel:
xi: float = config.XI_DEFAULT
model: pb.models.DixonColesGoalModel | None = None
teams: set[str] | None = None
def fit(self, df: pd.DataFrame, base_date: pd.Timestamp | None = None) -> "WorldCupModel":
base = base_date if base_date is not None else df["date"].max()
weights = pb.models.dixon_coles_weights(df["date"], xi=self.xi, base_date=base)
self.model = pb.models.DixonColesGoalModel(
df["home_score"].astype(int).to_numpy(),
df["away_score"].astype(int).to_numpy(),
df["home_team"].to_numpy(),
df["away_team"].to_numpy(),
weights=weights,
neutral_venue=df["neutral"].astype(int).to_numpy(),
)
self.model.fit()
self.teams = set(df["home_team"]) | set(df["away_team"])
return self
def predict_grid(self, home: str, away: str, neutral: bool = True):
return self.model.predict(home, away, max_goals=config.MAX_GOALS, neutral_venue=neutral)
def match_summary(self, home: str, away: str, neutral: bool = True) -> dict:
"""Full prediction payload for one fixture."""
g = self.predict_grid(home, away, neutral)
grid = np.asarray(g.grid)
ph, pd_, pa = (float(x) for x in g.home_draw_away)
# most-likely scorelines
flat = [(i, j, float(grid[i, j])) for i in range(grid.shape[0])
for j in range(grid.shape[1])]
flat.sort(key=lambda t: t[2], reverse=True)
top = [[i, j, round(p, 4)] for i, j, p in flat[:6]]
exp_home = float(sum(i * grid[i, :].sum() for i in range(grid.shape[0])))
exp_away = float(sum(j * grid[:, j].sum() for j in range(grid.shape[1])))
over25 = float(sum(grid[i, j] for i in range(grid.shape[0])
for j in range(grid.shape[1]) if i + j > 2.5))
btts = float(sum(grid[i, j] for i in range(1, grid.shape[0])
for j in range(1, grid.shape[1])))
return {
"home_team": home, "away_team": away, "neutral": neutral,
"p_home": round(ph, 4), "p_draw": round(pd_, 4), "p_away": round(pa, 4),
"exp_home": round(exp_home, 3), "exp_away": round(exp_away, 3),
"top_scores": top,
"grid": np.round(grid[:7, :7], 5).tolist(),
"over25": round(over25, 4), "btts": round(btts, 4),
}
def can_predict(self, *names: str) -> bool:
return self.teams is not None and all(n in self.teams for n in names)
def save(self, path=None):
path = path or (config.ARTIFACTS / "dixon_coles.pkl")
self.model.save(str(path))
(config.ARTIFACTS / "model_meta.json").write_text(
json.dumps({"xi": self.xi, "teams": sorted(self.teams)}, ensure_ascii=False),
encoding="utf-8",
)
@classmethod
def load(cls, xi: float = config.XI_DEFAULT) -> "WorldCupModel":
m = pb.models.DixonColesGoalModel.load(str(config.ARTIFACTS / "dixon_coles.pkl"))
meta = json.loads((config.ARTIFACTS / "model_meta.json").read_text(encoding="utf-8"))
obj = cls(xi=meta.get("xi", xi), model=m, teams=set(meta["teams"]))
return obj
if __name__ == "__main__":
df = load_training_frame()
print(f"training rows: {len(df)} ({df.date.min().date()} -> {df.date.max().date()})")
m = WorldCupModel().fit(df)
m.save()
print("Spain vs Brazil (neutral):")
s = m.match_summary("Spain", "Brazil")
print(f" 1X2 = {s['p_home']} / {s['p_draw']} / {s['p_away']} xG {s['exp_home']}-{s['exp_away']}")
print(f" top scores: {s['top_scores'][:3]}")