If you're an avid cricket fan, you may have heard about Monty, Fox Sports' cricket wicket predictor. It turns out Monty is now a little bit smarter, thanks to a tuneup of Monty's machine learning capabilities.
Monty entertained Fox Sports viewers last season by attempting to predict how likely it was that a wicket would fall in the summer test matches.
Monty works by using historical data from previous matches as well as live data, such as player behaviour and pitch conditions, to determine the chance of a wicket.
Monty was live but not broadcasting for the first Test series of the season and successfully predicted the first wicket of the game, in the Pakistani innings. It also predicted a wicket at the beginning of the first Australian innings—which was determined not out due to a no ball— as well as the actual first wicket of that same innings.
Fox Sports and Servian worked together to boost Monty's prediction power for the coming seasons. The results include a decline in the number of false positive wicket predictions and an improvement in overall accuracy.
"We managed to expand the number of machine learning features from 65 to 86, which allowed for more analysis and higher accuracy. The most significant addition was the bowler and batsman combinations," says Servian's Google practice lead Andrew Pym.
The features used in the first Test are based on the data provided to Fox Sports. Each ball is analysed by the model with a moving window, which looks to predict if a wicket is likely to fall within the next 10 minutes, based on the activity in the last five minutes.
In addition to the prediction, Monty also keeps track of the momentum of the bowling team.
As for the technical work, Pym explains that Servian used Google Cloud's serverless tools, including BigQuery and AutoML Tables.
"[We] added an event-driven approach to the serverless infrastructure. This supported the rapid development of multiple models for multiple games, which will be important for iterating and expanding Monty moving forward.
According to Servian, the design has lowered costs and improved start up times, with the use of Google Cloud AutoML Tables enabling 12 versions of the model to be developed in less than a week.
Fox Sports says it is now using Monty to create new, innovative ways to engage cricket fans this summer. Integration into broadcast and the Fox Sports apps ensures fans are alerted to key moments, including the fall of wickets, before they might occur.
Through this innovation, Monty's aim is to drive channel share and increase awareness of its core products such as the Fox Cricket app, the Foxtel Go app and live Test match coverage this summer.
