Machine Learning — What's Going On?

RateThatDog is always learning. We've built a lot of what we know on backtesting against a 5-year database of races, but backtesting can only take you so far — we're testing live, in the open, and the algorithms are always learning.

In the background, we're running a huge range of models: sprint-specific models, models for longer races, track-specific models, adaptive models that blend track and distance, and consensus models looking for high numbers of common selections across models.

Every day, the data comes in and the predictions for each race are made on the basis of the models that are running at the highest strike rate and the highest ROI. And with that data, the model keeps evolving.

Each of these models takes a blended approach: our performance, speed and suitability metrics, the history of races at this track, and the evidence we have from previous selections on performance from trap, in these race conditions.