تقسيم البيانات و Cross-Validation

from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(
    X, y,
    test_size=0.2,       # 20% للاختبار
    random_state=42,     # للتكرار
    stratify=y           # يحافظ على نفس نسبة الفئات
)

# تدريب
model.fit(X_train, y_train)

# تقييم على بيانات الاختبار (التي لم يرها النموذج)
score = model.score(X_test, y_test)

# أولاً: فصل test set
X_temp, X_test, y_temp, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# ثانياً: تقسيم الباقي لـ train و validation
X_train, X_val, y_train, y_val = train_test_split(X_temp, y_temp, test_size=0.25, random_state=42)
# 0.25 من 0.8 = 0.2 من الكل

# تدريب عدة نماذج
model_A.fit(X_train, y_train)
model_B.fit(X_train, y_train)
model_C.fit(X_train, y_train)

# اختيار الأفضل على validation
best_model = max([model_A, model_B, model_C],
                  key=lambda m: m.score(X_val, y_val))

# تقييم نهائي على test (مرة واحدة فقط!)
final_score = best_model.score(X_test, y_test)

K-Fold (K = 5):
              البيانات
   ┌─────┬─────┬─────┬─────┬─────┐
   │  1  │  2  │  3  │  4  │  5  │
   └─────┴─────┴─────┴─────┴─────┘

Iteration 1:  [Train] [Train] [Train] [Train] [Valid]
Iteration 2:  [Train] [Train] [Train] [Valid] [Train]
Iteration 3:  [Train] [Train] [Valid] [Train] [Train]
Iteration 4:  [Train] [Valid] [Train] [Train] [Train]
Iteration 5:  [Valid] [Train] [Train] [Train] [Train]

النتيجة: متوسط أداء 5 تجارب.

from sklearn.model_selection import cross_val_score

model = RandomForestClassifier(n_estimators=100, random_state=42)

# 5-fold CV
scores = cross_val_score(model, X, y, cv=5, scoring='accuracy')

print(f"دقّة كل fold: {scores}")
print(f"متوسّط الدقّة: {scores.mean():.3f}")
print(f"الانحراف المعياري: {scores.std():.3f}")
# متوسّط الدقّة: 0.847
# الانحراف المعياري: 0.023  → النموذج مستقرّ

scores = cross_val_score(model, X, y, cv=5)
print(f"std: {scores.std():.3f}")
# std: 0.15 — تقلّب كبير! النموذج غير مستقرّ.

cross_val_score(model, X, y, cv=5)

from sklearn.model_selection import StratifiedKFold

skf = StratifiedKFold(n_splits=5, shuffle=True, random_state=42)
scores = cross_val_score(model, X, y, cv=skf, scoring='accuracy')

from sklearn.model_selection import LeaveOneOut

loo = LeaveOneOut()
scores = cross_val_score(model, X, y, cv=loo)
# بطيء جدًا للبيانات الكبيرة

from sklearn.model_selection import TimeSeriesSplit

tscv = TimeSeriesSplit(n_splits=5)
for train_index, val_index in tscv.split(X):
    X_train, X_val = X[train_index], X[val_index]
    # تدريب النموذج

from sklearn.model_selection import GridSearchCV

param_grid = {
    'max_depth': [3, 5, 10, 20],
    'min_samples_split': [2, 5, 10],
}

grid_search = GridSearchCV(
    DecisionTreeClassifier(random_state=42),
    param_grid,
    cv=5,                    # 5-fold CV لكل توليفة
    scoring='accuracy',
    n_jobs=-1
)

grid_search.fit(X, y)
print(f"أفضل المعاملات: {grid_search.best_params_}")
print(f"أفضل دقّة CV: {grid_search.best_score_:.3f}")

from sklearn.model_selection import RandomizedSearchCV

random_search = RandomizedSearchCV(
    model, param_distributions, n_iter=20, cv=5, random_state=42
)

# ❌ خطأ: تطبيع قبل التقسيم
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)   # fit على كل البيانات!
X_train, X_test = train_test_split(X_scaled, y, test_size=0.2)

# ✅ صح: تطبيع بعد التقسيم
X_train, X_test = train_test_split(X, y, test_size=0.2)
scaler = StandardScaler()
X_train = scaler.fit_transform(X_train)   # fit فقط على train
X_test = scaler.transform(X_test)          # transform فقط على test

# ❌ خطأ: Feature selection قبل التقسيم
from sklearn.feature_selection import SelectKBest
selector = SelectKBest(k=10).fit(X, y)
X_selected = selector.transform(X)
# الآن train و test "يعرفان" معلومات عن بعض

# ✅ صح: Feature selection بعد التقسيم
X_train, X_test = train_test_split(X, y)
selector = SelectKBest(k=10).fit(X_train, y_train)  # fit فقط على train
X_train = selector.transform(X_train)
X_test = selector.transform(X_test)

from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression

pipeline = Pipeline([
    ('scaler', StandardScaler()),       # fit على train فقط
    ('model', LogisticRegression())     # fit على train فقط
])

# Cross-validation يستخدم الـ pipeline كوحدة واحدة
scores = cross_val_score(pipeline, X, y, cv=5)