Product Update: Hyper-Parameter Tuning & Corners Template

Product Notes from Gemini Head of Product Ally Brabant and Machine Learning Engineer Amod Sahasrabudhe

This week, Gemini Sports Analytics released two product updates: Hyper-Parameter Tuning and the Soccer Corners Template, both of which introduce new capabilities to our customers within the application!

Hyper-Parameter Tuning

On Tuesday, we launched Hyper-Parameter Tuning, our first feature available exclusively for our code-first customers!

What does this mean? Our Head of Product, Ally Brabant, explains:

One of Gemini’s core tenets is “to meet our customers where they are” and to match their workflows at all levels of technical knowledge. Our suite of APIs empowers customers to perform any task that can be done in the web-based application directly in their own code notebooks. Now, we have taken this a step further and created an endpoint to manually tweak a model to the customer's specifications.

This feature combines technology and expertise to allow subject-matter experts to go beyond autoML and fine-tune the model identified as the best one to meet their specific team needs. This will help produce even better results for teams looking to level up.

For example: You’re working on a soccer player archetype model for forwards and autoML returns a model with three types of forwards. This might be the “best” model, but if you, as the subject-matter expert, know that your team thinks there are actually four types of forwards, it would be hard to get buy-in from the coaching staff on the original model. With Gemini’s hyper-parameter tuning, you can tweak the model to only consider four groups of forwards, making it actually usable for your team.

Get in touch with us to learn more or for assistance in getting set up on our API suite!

Soccer Corners Template

In the second update this week, we made our Soccer Corners Template available to Gemini customers! Building on his research paper presented at MIT Sloan Sports Analytics Conference 2023 on Graph Neural Networks and their application in solving counterattacks, Gemini Machine Learning Engineer Amod Sahasrabudhe created this interactive tool to leverage event and tracking data to train a GNN to predict shot outcomes from corners.

Inspired by Google DeepMind and Liverpool Football Club's TacticsAI, which uses predictive and generative AI to provide elite coaches with insights on corner kicks, Amod built a front-end web app using data from our partners StatsBomb and SkillCorner that allows users to create custom corner scenarios, utilizing the GNN model to predict whether the tactical setup will lead to a shot. It also provides analysts and coaches an opportunity to dive deeper into designing their own set-piece plays.

Watch the video below to see how it works!

As always, we’d love to hear your feedback on these features, so please let us know if you’ve tried them and how you get on.

Watch this space for more product updates!