What is autopilot in machine learning?

Autopilot is a feature built into the DataRobot automated machine learning platform that allows anyone to develop machine learning models from a dataset. DataRobot’s Autopilot mode starts with a dataset and analyzes the dataset based on modeling decisions, such as the optimization metric. It then automates the feature engineering for lots of different types of features, including images. From there, it applies them to many different algorithms, resulting in thousands of combinations that are customized to your dataset. Finally, Autopilot starts searching for the best blueprint and then assembles the best individual solutions.

Autopilot Mode

What does Autopilot Mean?

Autopilot is a feature built into the DataRobot AI Platform that allows anyone, (even those without a deep understanding of mathematics, algorithms, or complicated programming languages), to develop machine learning models from a dataset.

To use DataRobot’s Autopilot, users upload a dataset, choose the “target” variable, and click the “Start” button.

machine learning autopilot

Why is Autopilot Important?

DataRobot’s Autopilot feature makes it possible for users of all analytical skill levels – not just data scientists – to build highly accurate machine learning models. Its built-in guardrails and data science best practices ensure users don’t miss critical steps in the modeling process, resulting in highly trustworthy models.

Autopilot makes it possible for users across an organization to uncover insights and make predictions using data, expanding the number of people in that can take part in artificial intelligence (AI) initiatives.

DataRobot + Autopilot

Once a user uploads a dataset, chooses a variable, and clicks “Start,” the DataRobot Autopilot kicks off a number of steps:

DataRobot calculates a Feature Importance metric, which measures the strength of the relationship between each feature and the target. This new metric is indicated by a green bar: the more full the bar, the more highly correlated that feature is with the target.
DataRobot automates many of the steps a data scientist would typically have to complete before building any models, such as creating a holdout set, specifying a validation approach, and selecting an accuracy metric. These steps not only save time, but also serve as guardrails for less sophisticated users to ensure they follow best practices.
DataRobot selects a list of model blueprints that are likely to perform well given the specific problem, casting a wide net in order to expose the modeling problem to a range of potential solutions. Blueprints are combinations of machine learning algorithms and a variety of preprocessing and feature engineering steps that boost model performance.

After DataRobot finishes building models, it ranks them by accuracy and offers a number of interpretability features that help explain and interpret the results.

Learn More About AI