Revise README for pmu_pipeline.py and remove stubspmu

[tiamosara81-tech]

pmu_pipeline.py

متطلبات: pip install requests pandas scikit-learn xgboost

import requests, pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.metrics import log_loss, brier_score_loss
import xgboost as xgb
import numpy as np

1) تحميل نتائج يومية عبر open-pmu-api (مثال)

def fetch_results(date_str):
# API مفتوح (مثال من GitHub project open-pmu-api)
url = f"https://open-pmu-api.vercel.app/results?date={date_str}"
r = requests.get(url, timeout=10)
r.raise_for_status()
return r.json()

2) تحويل JSON إلى DataFrame مبسّط

def build_dataframe(json_results):
rows = []
for race in json_results.get('races', []):
race_id = race.get('id') or race.get('raceId')
for pos, starter in enumerate(race.get('starters', []), start=1):
row = {
'race_id': race_id,
'horse': starter.get('horse_name'),
'jockey': starter.get('jockey'),
'trainer': starter.get('trainer'),
'draw': starter.get('draw'),
'weight': starter.get('weight'),
'fin_pos': starter.get('finishing_position') if starter.get('finishing_position') else 99,
'odds': float(starter.get('odds')) if starter.get('odds') else np.nan,
# هنا تضيف ميزات مشتقة لاحقاً
}
rows.append(row)
return pd.DataFrame(rows)

3) بناء ميزات بسيطة

def featurize(df):
# مثال: هدف = الفوز (fin_pos == 1)
df['target_win'] = (df['fin_pos'] == 1).astype(int)
# ميزات مبسطة: draw, weight, implied_odds
df['draw'] = pd.to_numeric(df['draw'], errors='coerce').fillna(0)
df['weight'] = pd.to_numeric(df['weight'], errors='coerce').fillna(df['weight'].median())
df['implied_prob_market'] = 1.0 / df['odds'].replace(0, np.nan)
df['implied_prob_market'].fillna(df['implied_prob_market'].median(), inplace=True)
# تحويلات سريعة للفارس/مدرب (count-based)
for col in ['jockey','trainer','horse']:
freq = df[col].value_counts().to_dict()
df[f'{col}_freq'] = df[col].map(freq).fillna(0)
return df

4) نموذج XGBoost لتنبؤ الاحتمال

def train_model(df):
features = ['draw','weight','implied_prob_market','jockey_freq','trainer_freq','horse_freq']
X = df[features].values
y = df['target_win'].values
X_train, X_test, y_train, y_test = train_test_split(X,y,test_size=0.2,random_state=42)
dtrain = xgb.DMatrix(X_train, label=y_train)
dtest = xgb.DMatrix(X_test, label=y_test)
params = {'objective':'binary:logistic','eval_metric':'logloss','seed':42}
bst = xgb.train(params, dtrain, num_boost_round=100, evals=[(dtest,'eval')], early_stopping_rounds=10, verbose_eval=False)
preds = bst.predict(dtest)
print("LogLoss:", log_loss(y_test, preds))
print("Brier:", brier_score_loss(y_test, preds))
return bst, features

Example flow

if name == "main":
date = "2025-10-31" # عدّل إلى التاريخ المطلوب
j = fetch_results(date)
df = build_dataframe(j)
df = featurize(df)
model, features = train_model(df)
# حفظ النموذج أو استعماله لاحقاً لتصنيف الاحتمالات لكل سباق