FLAML (Microsoft) Integration for Fast AutoML

FLAML (Microsoft) Integration for Fast AutoML

FLAML (Fast and Lightweight AutoML) from Microsoft Research is an AutoML library with a focus on minimal search time while maintaining high quality. It is used in Azure AutoML and Microsoft Fabric.

The key idea of FLAML

Cost-efficient search via cost-frugal Bayesian Optimization:

• Evaluates each configuration on a small fraction of the data (early stopping)
• Doesn't waste time on obviously bad configurations
• Adaptively redistributes the budget between algorithms

Basic examples

Classification:

from flaml import AutoML

automl = AutoML()
automl.fit(
    X_train, y_train,
    task='classification',
    time_budget=120,       # секунды
    metric='roc_auc',
    n_jobs=-1,
    eval_method='cv',
    n_splits=5,
    estimator_list=['lgbm', 'xgboost', 'rf', 'extra_tree']
)
print(f'Best model: {automl.best_estimator}')
print(f'Best AUC: {automl.best_result}')

Time series with FLAML:

automl = AutoML()
automl.fit(
    X_train, y_train,
    task='ts_forecast',
    time_budget=300,
    period=7,           # горизонт прогноза (7 дней)
    eval_method='holdout',
    estimator_list=['prophet', 'arima', 'lgbm', 'xgboost']
)

Integration with MLflow

from flaml import AutoML
import mlflow

def flaml_with_mlflow(X_train, y_train, X_test, y_test, run_name: str):
    with mlflow.start_run(run_name=run_name):
        automl = AutoML()
        automl.fit(X_train, y_train, task='classification', time_budget=300, metric='roc_auc')

        mlflow.log_param('best_estimator', automl.best_estimator)
        mlflow.log_param('best_config', str(automl.best_config))
        mlflow.log_metric('val_roc_auc', automl.best_result)

        # Тест на holdout
        from sklearn.metrics import roc_auc_score
        y_proba = automl.predict_proba(X_test)[:, 1]
        test_auc = roc_auc_score(y_test, y_proba)
        mlflow.log_metric('test_roc_auc', test_auc)

        mlflow.sklearn.log_model(automl, 'flaml_model')

    return automl

FLAML vs FLAML NLP/Vision: flaml[nlp] adds hyperparameter search for Transformer fine-tuning (HuggingFace). flaml[blendsearch] is an advanced BlendSearch strategy.

Timeframe: Basic FLAML integration – 1 day. MLflow tracking, custom estimators, parallel training, Azure AutoML pipeline – 1-2 weeks.