ml-training

ml-training

Production ML training: classifiers, regressors, and fine-tuned transformers. Covers dataset hygiene, feature engineering, cross-validation, hyperparameter tuning, and reproducible runs.

Dataset Hygiene

# Stratified train/val/test split — preserves class balance
from sklearn.model_selection import train_test_split
X_train, X_temp, y_train, y_temp = train_test_split(
    X, y, test_size=0.30, stratify=y, random_state=42
)
X_val, X_test, y_val, y_test = train_test_split(
    X_temp, y_temp, test_size=0.50, stratify=y_temp, random_state=42
)
# Result: 70/15/15 train/val/test

Always split BEFORE any feature engineering that uses target statistics (target encoding, target-aware imputation) — otherwise you leak test info into train.

# Compute class weights for imbalanced problems
from sklearn.utils.class_weight import compute_class_weight
weights = compute_class_weight('balanced', classes=np.unique(y_train), y=y_train)
class_weight_dict = dict(zip(np.unique(y_train), weights))

Feature Engineering

from sklearn.preprocessing import StandardScaler, OneHotEncoder
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline

numeric_cols = ['age', 'income', 'tenure_days']
categorical_cols = ['region', 'plan_tier']

pre = ColumnTransformer([
    ('num', StandardScaler(), numeric_cols),
    ('cat', OneHotEncoder(handle_unknown='ignore', sparse_output=False), categorical_cols),
])

# ALWAYS wrap preprocessing in Pipeline — fit on train only, transform on val/test
pipe = Pipeline([('pre', pre), ('clf', GradientBoostingClassifier())])
pipe.fit(X_train, y_train)
preds = pipe.predict(X_val)

Classifier Training (tabular)

import xgboost as xgb
from sklearn.metrics import roc_auc_score, average_precision_score, classification_report

model = xgb.XGBClassifier(
    n_estimators=500,
    max_depth=6,
    learning_rate=0.05,
    subsample=0.8,
    colsample_bytree=0.8,
    reg_alpha=0.1,
    reg_lambda=1.0,
    eval_metric='aucpr',
    early_stopping_rounds=50,
    random_state=42,
)
model.fit(X_train, y_train, eval_set=[(X_val, y_val)], verbose=25)

y_proba = model.predict_proba(X_test)[:, 1]
print(f"ROC-AUC: {roc_auc_score(y_test, y_proba):.3f}")
print(f"PR-AUC : {average_precision_score(y_test, y_proba):.3f}")