automl

AutoML automates machine learning pipeline components: feature engineering, model selection, and hyperparameter tuning. **Scope**: Broader than NAS - covers entire ML workflow, not just architecture. **Components**: **Feature engineering**: Automatic feature selection, transformation, creation. **Model selection**: Choose among algorithms (random forest, neural net, XGBoost). **Hyperparameter optimization**: Find best hyperparameters automatically. **Pipeline integration**: Combine preprocessing, model, postprocessing. **Tools**: **Google AutoML**: Cloud service for custom models. **Auto-sklearn**: Automated scikit-learn. **H2O AutoML**: Open source platform. **AutoGluon**: Amazon, strong tabular performance. **FLAML**: Microsoft, fast lightweight. **For deep learning**: Primarily hyperparameter tuning and NAS. Less automation of feature engineering (learned by network). **Benefits**: Democratizes ML (non-experts can build models), saves time, may find better configurations than manual tuning. **Limitations**: Compute cost, may not understand domain constraints, black box models, limited customization. **Current use**: Very common for tabular data, hyperparameter tuning. Enterprise ML platforms increasingly include AutoML.