6 AutoML with a fully customized search space

This chapter introduces customization of the entire AutoML search space in a layer-wise fashion without wiring up AutoML blocks, giving you more flexibility in designing the search space for tuning unsupervised learning models and optimization algorithm. You will also learn how to control the model training and evaluation process to conduct a joint tuning and selection of deep learning models and shallow models. This allows you to tune models with different training and evaluation procedures implemented with different ML libraries.

6.1 Customizing the search space in a layer-wise fashion

In chapter 5, you learned how to perform hyperparameter tuning and model selection by specifying the search space with AutoML blocks. You also know how to create your own AutoML blocks if the built-in blocks do not fit your needs. However, there may be some scenarios that are hard to address by wiring together AutoML blocks or where this simply doesn’t seem to be the best approach.

6.1.1 Tuning an MLP for regression with Keras Tuner

6.1.2 Tuning an autoencoder model for unsupervised learning

6.2 Tuning shallow models with different search methods

6.2.1 Selecting and tuning shallow models

6.2.2 Tuning a shallow model pipeline

6.2.3 Trying out different search methods

6.3 Controlling the AutoML process by customizing tuners

6.3.1 Creating a tuner for tuning scikit-learn models

6.3.2 Creating a tuner for tuning Keras models

6.3.3 Jointly tuning and selection among deep learning and shallow models

6.3.4 Hyperparameter tuning beyond Keras and scikit-learn models

6.4 Summary