Today we’re pleased to announce an exciting new partnership for data scientists and machine learning practitioners everywhere: The integration between Modelbit and Eppo allows ML practitioners to A/B test their running models using Eppo!
Machine Learning is an inherently experimental discipline. Data scientists and ML practitioners put a premium on the ability to quickly test different features, model types, ML frameworks, and even hyperparameters. This need only accelerates once the model hits production: New data arrives every day, new modeling technologies are invented, and old models tend to drift. The most mature machine learning teams are always A/B testing new model versions to achieve peak performance.
Eppo makes it easy to run impactful A/B tests and provides a suite of tools for managing these tests and analyzing the results. Modelbit provides a platform for data scientists to rapidly deploy ML models and manage all their running models. It’s a natural fit: By integrating Eppo and Modelbit, we’re able to provide best-in-class A/B testing tools alongside the best-in-class model deployment platform.
Inside of Modelbit, find the Eppo integration inside the “Integrations” tab in “Settings” and click to get started.
From there, simply pop in an Eppo API key:
Make sure your API key has at least “Read” access for “Feature Flagging.”
To get started with your newly-created Eppo+Modelbit integration, head over to Eppo and create a new experiment. Ensure that your experiment has multiple variations.
Next, head back to Modelbit. Click the “Endpoints” tab and make a new Request Splitting endpoint. Choose “Using Eppo” for the Split Assignment, and enter your Eppo experiment name and the supplied key you’d like to split on:
And the rest is history! Modelbit will automatically split traffic between your different models, and Eppo will manage traffic allocations and track results.
It’s still early days for machine learning and experiment-driven product development. As new machine learning technologies are created every day, it becomes increasingly important to apply a thorough, results-driven approach to making improvements. The most mature data organizations are rigorously A/B testing new models before making changes in production. We expect this trend to continue – and we are committed to supporting these customers every step of the way!