Productionalizing H2O Driverless AI Models

Training a good machine learning model is an extremely difficult process. Good data science practitioners must first determine if the data they have is useful at all. Next, do they have to cleanse or munge the data to put it into the proper format for the machine learning algorithm they are planning to use? Then, you might need to create new features based off the original data that provide better signal for predicting the target value, and consider what hyperparameters to use when training the algorithm. To name a few steps. However, this is only the first step in creating a useful model. The next step, and one that is arguably just as important is productionalizing a model. In many cases, companies have strict rules about how a model must behave or in what kind of infrastructure a model must run in production. As an example, some companies require only Java models, and data scientists who produced the model in R or Python must then pass their code to a data engineer who will take a month or two to translate the model from the original to Java. This kind of restriction is often times the major barrier to entry when it comes to pushing new machine learning models to production. Join our webinar to learn about common approaches to productionalizing models, and how to apply these practices to models produced by H2O Driverless AI. Join our webinar to learn: • Some common challenges associated with productionalizing models in different infrastructures • Good practices when productionalizing models, specifically related to models produced by Driverless AI • Some generic examples of how to productionalize a model • Time permitting: a live coding exercise to productionalize a Driverless AI Mojo