Exploring the DataRobot AI Platform

In this chapter, we will turn our focus to the DataRobot AI platform, a paid software platform that provides a powerful toolkit for deep learning use cases. DataRobot allows its users to streamline the complex stages of the machine learning life cycle. It presents an intuitive interface for data scientists, engineers, and researchers who wish to harness the power of machine learning for their projects and businesses. As we delve into the workings of DataRobot, you will learn how it simplifies and accelerates the creation, training, deployment, and government of intricate deep learning models. Thanks to features designed for automation and ease of use, it empowers users to focus on what truly matters—extracting significant value from their machine learning applications.

Our exploration will highlight the key functionalities of DataRobot, underlining its potential as a catalyst in the evolution of deep learning solutions. DataRobot aspires...

Technical requirements

We will have a practical topic in this chapter to make predictions using a DataRobot deployed model. We will be using Python 3.10 and we will require the following Python libraries to be installed:

• datarobotx==0.1.17
• pandas==2.0.3

The code files are available on GitHub at https://github.com/PacktPublishing/The-Deep-Learning-Architect-Handbook/tree/main/CHAPTER_18, and the dataset can be downloaded from https://www.kaggle.com/datasets/dicksonchin93/datarobot-compatible-house-pricing-dataset.

Additionally, a paid or free trial account is needed to access DataRobot. To subscribe for a trial account, do the following:

1. Visit the DataRobot website at https://www.datarobot.com/trial/.
2. Fill up your credentials under the Start For Free interface on the right side of the web page and click on the Submit button.

A high-level look into what the DataRobot AI platform provides

The DataRobot AI platform provides data ingestion, data preparation, data insights, model development, model evaluation, model insights and analysis, model deployment, and model governance through model monitoring and model maintenance tools that work seamlessly with each other. While DataRobot streamlines the deep learning life cycle, it is important to note that the planning stage still requires human input to define the goals and scope of the project. Additionally, you are still required to consume the insights, reports, and results made easy for you to obtain. Ultimately, this means that such a platform is a tool that can assist any machine learning practitioner instead of being a replacement for data scientists, machine learning engineers, machine learning researchers, or data analysts. Think of AI platforms such as DataRobot as being powerful calculators that can help you solve complex math problems quickly and accurately...

Preparing data with DataRobot

The first part of what the platform offers is the data preparation component. DataRobot simplifies the data preparation process by offering a range of features to streamline data ingest, cleaning, transformation, and integration. Let’s dive into these features in detail.

Ingesting data for deep learning model development

The development of deep learning models in DataRobot begins with the pivotal step of data ingestion. This process allows you to directly import your data from various sources, including cloud storage (such as AWS S3), Google Cloud Storage, local files, or databases such as PostgreSQL, Oracle, and SQL Server. The platform accepts diverse file formats, including CSV, XLSX, and ZIP files. Additionally, the platform supports image, text, document, geospatial, numerical, categorical, and summarized categorical data through secondary datasets as input data types. For the target data types, the platform supports numerical, categorical...

Executing modeling experiments with DataRobot

DataRobot currently provides two ways to execute modeling experiments: DataRobot Classic and Workbench. Workbench is where an experiment will be managed under a use case, focusing on extracting value from a use case more seamlessly, and DataRobot Classic is the original AutoML experience where a modeling experiment is called a project. A project, or a modeling experiment here, encompasses the same components, which include modeling machine learning, gathering model insights and prediction insights, and making one-off batch predictions. We will dive deeper into these three components.

Deep learning modeling

DataRobot provides modeling configurations and tasks in the form of directed acyclic graphs (DAG) called blueprints. The individual nodes in the graph are grouped up into the following:

• Input data: The input nodes can be any of the supported input data types.
• Data preprocessing tasks: They consist of data regularization...

Deploying a deep learning blueprint

DataRobot allows the deployment of a model directly through a trained blueprint in an experiment or a project, which we will explore in the next practical section. However, for more advanced users, the platform also allows the deployment of custom models through the Custom Model Workshop feature. Custom inference models are user-created, pre-trained models that can be uploaded to DataRobot as a collection of files coupled with either a drop-in environment or by a requirements.txt file. Once uploaded, users can create, test, and deploy custom inference models to DataRobot’s centralized deployment hub. These custom models support different model types, which include regression, classification, and unstructured types where the input and output can be of various types.

To ensure the reliability and compatibility of your custom models, DataRobot provides a comprehensive testing suite in the Custom Model Workshop. The custom model testing suite...

Governing a deployed deep learning blueprint

In this section, we will discuss how DataRobot enables users to govern their deep-learning models effectively by providing comprehensive tools for model utilization, monitoring, and maintenance. With a focus on seamless integration, DataRobot allows users to deploy AI applications on cloud-based or on-premises infrastructure, manage prediction outputs, and monitor model performance using custom metrics and alerts. Furthermore, the platform supports data drift detection and offers retraining capabilities for continuous model improvement. We will explore these features in detail, demonstrating how DataRobot empowers users to efficiently manage their deep learning models and ensure optimal performance throughout their life cycle.

Governing through model utilization in DataRobot

Users can access their models through various means, such as API calls, Python interfaces, or DataRobot-made applications called AI Apps. The platform supports...

Exploring some customer success stories

DataRobot has empowered numerous organizations to achieve remarkable success through the implementation of deep learning solutions, particularly in handling unstructured data such as text and images. While most of these success stories remain confidential, we are fortunate to have a few customers who have enthusiastically shared their inspiring experiences, showcasing the transformative potential of deep learning in various industries. Some of these notable successes include the following:

• Lenovo, a leading technology company, successfully implemented DataRobot’s Visual AI in its Brazilian laptop manufacturing facility to improve quality control and increase productivity. The Visual AI system helped increase label verification accuracy from 93% to 98% by automating the comparison of identification labels on laptops with their respective bill of materials. This implementation not only reduced errors in the manual labeling process...

Summary

This chapter explored the DataRobot AI Platform and showcased the benefits an AI platform can provide to you in general. DataRobot streamlines the complex stages of the machine learning life cycle, providing an intuitive interface for data scientists, engineers, and researchers. By harnessing the potential of AI platforms such as DataRobot, users can accelerate the creation, training, deployment, and governance of intricate deep learning models, focusing on extracting significant value from their machine learning applications.

DataRobot offers automation, collaboration, and scalability for machine learning use cases. DataRobot provides support for various data types and advanced features such as bias and fairness mitigation, Composable ML, custom tasks, advanced tuning, and time-series modeling. DataRobot also enables users to deploy AI applications on cloud-based or on-premises infrastructure, manage prediction outputs, monitor model performance, and maintain models implemented...