Book Image

Agile Machine Learning with DataRobot

By : Bipin Chadha, Sylvester Juwe
Book Image

Agile Machine Learning with DataRobot

By: Bipin Chadha, Sylvester Juwe

Overview of this book

DataRobot enables data science teams to become more efficient and productive. This book helps you to address machine learning (ML) challenges with DataRobot's enterprise platform, enabling you to extract business value from data and rapidly create commercial impact for your organization. You'll begin by learning how to use DataRobot's features to perform data prep and cleansing tasks automatically. The book then covers best practices for building and deploying ML models, along with challenges faced while scaling them to handle complex business problems. Moving on, you'll perform exploratory data analysis (EDA) tasks to prepare your data to build ML models and ways to interpret results. You'll also discover how to analyze the model's predictions and turn them into actionable insights for business users. Next, you'll create model documentation for internal as well as compliance purposes and learn how the model gets deployed as an API. In addition, you'll find out how to operationalize and monitor the model's performance. Finally, you'll work with examples on time series forecasting, NLP, image processing, MLOps, and more using advanced DataRobot capabilities. By the end of this book, you'll have learned to use DataRobot's AutoML and MLOps features to scale ML model building by avoiding repetitive tasks and common errors.
Table of Contents (19 chapters)
1
Section 1: Foundations
5
Section 2: Full ML Life Cycle with DataRobot: Concept to Value
11
Section 3: Advanced Topics

Defining the ML problem

Before we get into the details of ML, it is important to note that not all problems are appropriate to be solved with ML. For a problem to be a good candidate, it should have the following characteristics (we will focus only on supervised learning (SL) problems for now):

  • There is a clear target or label value that would be useful if an algorithm can predict it. In the absence of an algorithm this value remains unknown, requires a person's judgment, or requires substantial effort for it to be determined. Sometimes, the target will not be the actual variable of interest but a critical component of that calculation. This part is not always obvious, but the problem analysis you did in previous sections of this chapter will certainly help in clarifying which variable makes the best target.
  • You have access to a large enough historical dataset that contains the values of the target or label you wish to predict. You will need to create a list of data...