This is just a proof of concept right now, but I basically want to create a little dashboard in ther browser that will follow live baseball games and, in real-time, show the probabilities of each outcome of each player's next at-bat. Then you can look at the numbers of your chart for who's about to be up and easily to compare it to bets available at your sportsbook, and this should then allow you to always bet at a theoretical advantage.
Bets Available
This webpage has a nice description of various baseball in-game bets that are available. There is normally a "money-line" type bet on whether the next at-bat will be a hit (for example, the line would be expressed at +300, meaning you bet $100 to win $300 if it's a single or better). There are sometimes other options to bet on for the at-bat outcome such as betting on a strikeout, fly ball, or walk as well as being able to specify what type of hit. Another thing to note is that this type of sports betting (and really sports betting in general) just recently became legal in the new jersey and new york area, meaning that it's still in its infancy and may be exposing some ripe dollar-fruit to be picked off by sharks.
Data Source
It's very convenient that MySportsFeeds exposes a free api (including a real-time api!) that we can use as the data source for the application and possibly not even need our own backend server. MySportsFeeds provides many different real-time streams that can be useful in predicting the probability of a hit such as: Scoreboard, Game BoxScore, Player Game Logs, Team Game Logs, Active Players, and Daily Player Stats. Of course open source is not always the most accurate / reliable, and we should give an open mind long term on paid data sources if this starts to become a legit revenue-generating thing.
The Algorithm
I am very excited that tensorflow now has a JavaScript library, and this example is extremely similar to what I want to do here (although the example is in python so please excuse the ugly underscore-happy syntax 😂😂). Our baseball predictor is trying to solve a regression problem, well many regression problems, as it simultaneously predicts the probabilities of various at-bat outcomes for every player's next at bat, on both teams, continuously! In theory we should be able to just feed in the data from MySportsFeeds and so what it predicts!
Ok Then
This was all just an idea I had, but actually writing out this blog post has really made me realize that this is definitely a doable project and could potentially work super well. I have to make a tough decision in terms of React vs Angular here because I've been using React for a while, but with all the streaming and data subscriptions maybe the observable-based Angular would be cleaner and easier to implement. With me luck, and let me know if you are working on anything similar! 😉
RSS Feed