Today I had been browsing Internet looking for some information about machine learning and betfair bots. I found Betfair Australia web site:
https://www.betfair.com.au/hub/
The web site presents best trading tools for betfair and polarizes machine learning and automation of different battings strategies.
Some data scientists even offer their models:
https://www.betfair.com.au/hub/tools/models/
For public using so anyone is able to access their data and used them free of charge.
Betfair staff created set of tutorial articles to automate their ML strategies by best available tools for automations:
https://betfair-datascientists.github.io/#using-third-party-tools-for-automation
“I have a set of tips that I've taken from our DataScientists' Racing Prediction Model, but this approach should work for any set of tips have. My goal is to create an automated process which will let me choose my tips for the day, then walk away and the program do the leg work.”
So the aim of automation is to bet on first or second selection ranked by betfair machine learning (ML) model for horse racing:
https://www.betfair.com.au/hub/racing-tips/
The problem is that suggested automation actually does not take ranked selections from betfair ML model, but places bets on first or second selection in order betfair loads the market.
“- Choosing your selections
The final step is to choose which selections you want to bet on. In this example I just chose the number 1 selection chosen by the Data Scientists in their Racing Prediction Model. Just click on the dropdown in the 'Automation Nomination Selection 1' column for each race and choose your selection.”
Bfexplorer Automation
I took some time and made my own automation using betfair ML models for horse racing and greyhounds racing.
Bfexplorer approach to automate betting or trading utilizing data freely available on the Internet is based on creating “data providers” for particular strategy, so in this case I created code to load the web pages:
https://www.betfair.com.au/hub/racing-tips/
https://www.betfair.com.au/hub/greyhound-ratings-model/
And take relevant information, so for horse racing the rank of selections and for greyhounds racing the selection rated prices.
Betfair staff suggest: “This makes for a bit of prep work, copying the list of runners and their rating into an Excel spreadsheet. As a minimum you'll need a list of runner names (including the runner number followed by a full stop, i.e. 1. Runner Name) in one column and their rating in another in an Excel sheet.”
https://betfair-datascientists.github.io/thirdPartyTools/cymaticTraderRatingsAutomation/
But of course in my case this procedure is really fully automated.
The next step is to create “trigger bot”, so code responsible for synchronizing data with betfair ones, and evaluation of data to return a list of market selections by different criteria.
In this case for greyhound racing ML we have got rated price, so the price/odds calculated by ML, which are influenced by sectional time data, box number, market prices and many other variables.
So the rated price/odds expresses probability of greyhound to won a race (probability = 1.0 / odds).
Betfair staff suggests using:
“In short, I want to back runners when:
the available to back price is better than the rating for that runner by a variable percentage
they have a rating less than 5
the scheduled event start time is less than 2 minutes away
the event isn't in play”
In my case I want to check ML predictability so I just use price different and/or rated price, and created set of evaluating rules by: Best Rated, Best Selection, Close To Rated, Positive Close To Rated and Negative Close To Rated.
A trigger bot code makes selection process, but exact betting or trading strategy is executed by the action bot (strategy/bot defined by parameter BotName), and because all bfexplorer bots use parameters to describe selection process or your exact strategy, then even when your bot code is built by another person, your know-how is hidden.
You don’t have to worry thata someone will steal your strategy you work for years on.
Another advantage of such trigger bot strategy is that you can trigger/execute any strategy this way. For instance your action bot can place back or lay bet, execute trading strategy back/lay or lay/back for set profit or loss target, or even you can dutch your dedicated selections.
So in testing you can create set of action bots strategies, and set of trigger bot strategies, just by setting different parameters. Then you can test simultaneously even more than 20 strategies, and then choose only best profitable ones to for couple weeks of trial to select the best strategy.
When starting to test your strategy you create its settings in “My Strategies to Execute”, and then manually execute it on some markets to see in the “Bet Event” view how the strategy works on betfair markets.
The final step is to use your strategy fully automatically by using “Strategy Executor” tool.
If you are my subscriber and want to test this betfair ML strategy, send me your request through support web page.