Book Image

Practical Automated Machine Learning Using H2O.ai

By : Salil Ajgaonkar
Book Image

Practical Automated Machine Learning Using H2O.ai

By: Salil Ajgaonkar

Overview of this book

With the huge amount of data being generated over the internet and the benefits that Machine Learning (ML) predictions bring to businesses, ML implementation has become a low-hanging fruit that everyone is striving for. The complex mathematics behind it, however, can be discouraging for a lot of users. This is where H2O comes in – it automates various repetitive steps, and this encapsulation helps developers focus on results rather than handling complexities. You’ll begin by understanding how H2O’s AutoML simplifies the implementation of ML by providing a simple, easy-to-use interface to train and use ML models. Next, you’ll see how AutoML automates the entire process of training multiple models, optimizing their hyperparameters, as well as explaining their performance. As you advance, you’ll find out how to leverage a Plain Old Java Object (POJO) and Model Object, Optimized (MOJO) to deploy your models to production. Throughout this book, you’ll take a hands-on approach to implementation using H2O that’ll enable you to set up your ML systems in no time. By the end of this H2O book, you’ll be able to train and use your ML models using H2O AutoML, right from experimentation all the way to production without a single need to understand complex statistics or data science.
Table of Contents (19 chapters)
1
Part 1 H2O AutoML Basics
4
Part 2 H2O AutoML Deep Dive
10
Part 3 H2O AutoML Advanced Implementation and Productization

Summary

In this chapter, we learned about some of the optional parameters that are available to us in H2O AutoML. We started by understanding what imbalanced classes in a dataset are and how they can cause trouble when training models. Then, we understood oversampling and undersampling, which we can use to tackle this. After that, we learned how H2O AutoML provides parameters for us to control the sampling techniques so that we can handle imbalanced classes in datasets.

After that, we understood another concept, called early stopping. We understood how overtraining can lead to an overfitted ML model that performs very poorly against unseen new data. We also learned that early stopping is a method that we can use to stop model training once we start noticing that the model has started overfitting by monitoring the performance of the model against the validation dataset. We then learned about the various parameters that H2O AutoML has that we can use to automatically stop model training...