"Bring Your Own Model" is a Data Scientist's Most Preferred Functionality

What is a data scientist’s worst nightmare? Well, it’s not getting low model accuracy or ranking low in a Kaggle competition. In fact, it’s developing a cutting-edge AI model that can’t operationalize on an analytic platform. Unfortunately, this is the reality for many data scientists. This happens when data scientists create their AI model in their tool of choice, but the model is not compatible with the operationalization platform.  

Such a situation would lead to a huge loss of work as well as a lost business opportunity. With so many programming languages and AI tools in the market, the gap between tools and operational platforms is constantly expanding. But as proven time and time again—programming languages and AI tools are imperative. The freedom to use algorithms which produce robust, accurate models and provide a solid, quantifiable ROI is an essential part of data science work.  

So how do you provide the freedom of choice for tools and ensure model interoperability for operationalization purposes? Teradata has the answer—Bring Your Own Model (BYOM). With one of Teradata’s most differentiating features, users are free to develop models using the tools of their choice and import them into Teradata Vantage for operationalization.  

“BYOM is a model sharing capability that allows users to operationalize externally created AI models inside Teradata Vantage.” 

This feature continues to support Teradata’s strategic analytics framework that gives data-driven enterprises a step-by-step solution for deploying analytical models at scale. With BYOM, businesses quickly realize a greater return on investment. Plus, BYOM provides ease in developing analytical models through increased model operationalization, expanded analytic use cases, and a streamlined approach to data-driven decision-making.  

BYOM within Teradata’s Recommended Framework 

Here at Teradata, we have a recommended framework to plug the proof-of-concept to production gap. By following the steps to prepare the data, train models, and deploy models, enterprises can increase their business value and ROI. The BYOM functionality greatly improves this process, as indicated in bold within the framework. 

This framework establishes a straightforward roadmap for businesses to create robust, efficient, and easily deployed processes that ensure AI/ML projects live up to their promise and deliver business value.   

Breaking it down, the framework consists of three steps: Data Preparation, Model Training, and Deployment. The first section, Data Preparation, is exactly what it sounds like – users preparing and managing their data at any scale. The core features within the data are then extracted and used for analytic modeling. Once these features are created, they are curated within an Enterprise Feature Store for easy access in the future. The next step is Model Training, which begins with creating models from the features delivered in the first step. For model functionality, users can choose either native adaptation within Vantage or BYOM. In either case, ClearScape Analytics ensures that a wide range of models, often used in combination, are made available for operationalization. Lastly, we have the Deployment step. With Vantage’s ModelOps service, users can manage end-to-end analytic model creation at scale. Vantage will monitor model performance and automatically trigger rescoring or model updates, all while maintaining model, features, code, and data lineage.   

BYOM further supports the recommended framework as it engages with a large partner network to enable better analytic outcomes through the full analytics lifecycle. Data scientists can leverage the power of ClearScape Analytics through their tool of choice. Plus, ClearScape Analytics provides seamless integration with commonly used machine learning tools.   

Benefits to BYOM 

Although there are many benefits to BYOM, there are three that are especially important: reduced risk, flexibility, and partnerships.  

BYOM reduces risk in model creation and deployment. The functionality ensures customers can retain their investments in model development technologies without any risk or functionality loss when deployed in Vantage. This is realized by importing externally created predictive models through open-source packages or third-party solutions into Vantage, then scoring these models in parallel. This is run using all the data available to the user.  

BYOM provides flexibility to the users. Teradata data scientists can use any of their preferred open-source tools to be executed in parallel to native Vantage analytic functions. This enables the operationalization of insights without the need to sample data or create data silos outside of Vantage.  

Teradata Vantage partners with a variety of leading advanced analytics and AI/ML vendors. Our framework naturally incorporates existing and new partners in Teradata’s analytics and AI/ML portfolio, providing customers and the data science community with the industry’s most optimal analytics ecosystem. These partnerships provide business users with answers and insights in minutes, enable more robust models for deeper insights, and deliver faster model refresh, updates, and replacements. With the BYOM functionality, Vantage can address the most stubborn challenges facing organizations and allow them to quickly realize value from their AI/ML investments.  

BYOM with Various Model Formats and Partner Tools 

BYOM can be executed for PMML, ONNX, H2O, and Dataiku models. The BYOM functionalities empower users to accelerate digital transformations and bring the power of AI to everyone through a very intuitive user experience. All users are guaranteed a high-quality experience with speed, scalability, and reliability.    

Data scientists can convert trained ML models to common serialization formats like PMML and ONNX, or retain models in H2O MOJO and Dataiku formats. ClearScape Analytics stores standard PMML, ONNX, H2O MOJO, and Dataiku models in Vantage Analytic Database tables and utilizes BYOM software packages to score and predict stored models.   

There are many benefits of BYOM with any of these formats, starting with operationalizing and realizing business value from Vantage ClearScape Analytics. The BYOM feature is $0, which means there is no purchase or license cost. BYOM retains its simple user experience with no coding required while leveraging Vantage’s strengths including Performance and Advanced Analytics at Scale. Plus, BYOM incorporates “Production Quality” analytics scoring capability and compliments Script Table Operator by expanding scoring use cases beyond R and Python models.  

Overall, ClearScape Analytics, and BYOM, continue to expand its capabilities and usability, making it one of the most preferred data scientist functionalities!