Introducing The Pena.lt/y Web App

Posted by Martin Eastwood April 11, 2014 10 Comments 1819 views

I’ve been publishing various football odds and predictions on the site here since its inception so I thought it was time to get  a bit more organised about it.

I’d started refactoring all my code recently to automate it so that the models automatically spotted upcoming fixtures and predicted the odds without me needed to tell them to. This all worked fine except that I had to poke around in a database to look at the results so I started developing an interface for it to avoid the tedium that it SQL.

Around this time @juzbo from sportshacker.net challenged me to see who could develop a web app first and and so due to my inherent competitiveness the pena.lt/y web app was born.

The pena.lt/y web app is essentially a website connected to a computer programme running in the cloud that monitors upcoming football fixtures and models their expected outcomes and probabilities.  As well as individual fixtures the pena.lt/y web app also models full seasons so you can see how likely teams are to win the league, reach Europe or get relegated etc. Plus you can see their Pythagorean Predictions too.

Whenever you view a webpage on the pena.lt/y web app section of the site it talks to this programme running the modelling via an API to get the latest fixtures and probabilities which it then uses to dynamically create the page for you to look at. This all means that everything is now fully automated and I don’t need to do anything, the model runs itself and the website updates all its pages and images etc without me having to do anything.

Plus, because the pena.lt/y web app is powered via an API other software can hook up to it too. For example I currently have a bot sniffing around Betfair that checks the odds against mine to see where the value lies.

If you are interested in seeing more then take a look here or via the ‘Fixture Predictions’ header in the menu at the top of the page.

Anyway, let me know what you think – what works, what doesn’t and what I should add next!

 

About Martin Eastwood

Martin is football fan and data scientist. In his spare time he likes to combine the two and write about the mathematical analysis of football.

View all post by Martin Eastwood

There are 10 Comments

  1. Jon
    - April 11, 2014
      -   Reply

    API please? Account is xxxxxxxxc

    • Martin Eastwood
      - April 18, 2014
        -   Reply

      Still doing a bit more work on the api so not quite ready yet

  2. - April 18, 2014
      -   Reply

    Fantastic work. I’m not sure thought where I can see the Pythagorean Predictions in the app?

    • Martin Eastwood
      - April 18, 2014
        -   Reply

      Go to the competition menu at the top, choose the league you want then its under the pythag tab

      • - April 19, 2014
          -   Reply

        Ah, there it is. Thanks again for the app,

  3. James
    - April 21, 2014
      -   Reply

    Fantastic work Martin and thanks for making your work public.

    I’m interested to know what inputs go into your model as i’ve been trying to incorporate shot data into a Dixon & Coles model but i’m not sure which route do go down.

    Also, you mention you have a “bot sniffing around betfair” to see “where the value lies”, what is your criteria for selecting bets because from what i’ve seen (small sample size) your model is less accurate than the bookmakers (as would be expected).

    • Martin Eastwood
      - April 21, 2014
        -   Reply

      Hi James,

      I’d recommend starting off with the basic Dixon and Coles model as described in their paper and tweak from there. Honestly, their model takes a lot to beat and even then the gains are fairly small.

      The Betfair bot is more of a toy project rather than anything serious, there is generally poor liquidity in the Betfair markets especially for Asian Handicaps. At the moment it is taking the simplistic approach of looking at the ratio between their odds and mine and when it exceeds a certain threshold it flags it up to me but with Betfair’s liquidity this is an increasingly rare event..

      I’m interested in your thoughts about the accuracy. I track accuracy using rank probability scores and it seems roughly comparable – sometime the bookmakers are better, sometimes my model is. If you see anything that stands out as being obviously incorrect then do let me know!

  4. James
    - April 25, 2014
      -   Reply

    I also track accuracy using rank probability scores – I wonder where I could have got that idea from!

    I’ll admit to having only calculated the RPS for the 1×2 market, I would have thought that accurate predictions for over/unders and asian handicaps would require some adjustments for ‘in-play’ events (like looking at performance for different game states) but maybe I’m wrong.

    Your model is certainly comparable with the bookmakers, which is why I was wondering if there was a specific subset you were selecting for bets. ie. if your model ‘exaggerated’ something specific, draws for example due to a rho adjustment.

    I’ve only just got ‘a’ Dixon & Coles model running in R, no idea how to add the weighting function in yet though. And I’ll take any fairly small gain I can find.

  5. Nuno
    - May 9, 2014
      -   Reply

    Ei Martin. excellent work. I was wondering how i could get good results by betting with your advices. Should i bet always on your favourite outcomes? Should i bet when the bet is a value bet even when the outcome isn’t your favourite (i.e outcome with the higher probability).

    Thanks in advance, keep up with the good work. it’s inspiring for everyone

    • Martin Eastwood
      - May 10, 2014
        -   Reply

      There are lots of different strategies out there but going with the value and using the kelly criteria is a good place to start

Write Your Comment

  • fede

    oh too bad you havent time :( i was following with …

  • Martin Eastwood

    The app page has gone for the time being but the p …

  • fede

    First of all, thanks for your hard and very intere …

  • Martin Eastwood

    Sorry Justin, I don't have a spreadsheet with it i …

  • Martin Eastwood

    Cool, will take a look! …

Latest Tweets

  • Loading...