After reviewing some of the discussion surrounding the legend *psychoff*, it occurred to me (as it does many) that this individual has a rare gift for reading the game. This is a soft skill and although it can be taught, a lot of it is pattern recognition and as such can potentially be modestly modelled in software.
I am in no way suggesting that years of experience and market exposure can be encapsulated in a simple algo. What I'm getting at is that this soft skill is a collection of small tangible units that are called into play during the decision making process and assembled together in a unique configuration so as to make the most balanced decision for a given vector within the game. One such vector may simply be to sit on your hands.
To this end, IPT (Scanner) came into my head. I had used this tool back in 2017 and although i never really exploited its full potential, I became aware of the mass of data that it was emitting for each moment of the game. This got me thinking (as a starting point) that many of these data points could be viewed as assembly components and thus coupled together to leverage documented scenarios that occur week in week out in most football games.
Now I know that this tool is in wide spread use and therefore one could expect it's currency value to be diminishing, but, I think that potentially there are many hidden data assembly components that can/could be extractd from such a data rich tool and fed back into a rules engine and onto BA for bet processing.
To this end, I wanted to open a discussion on the strengths and weaknesses of taking a *blinds down, exclusively data driven approach* to finding value within the wide range of markets that football presents. This may simply be about looking at data points A&B and entering the frey - or it could be something more exotic like looking at data points A&B but only if datapoints C,D and E fell within x minutes of the former.. /etc, etc.
In short, it's going to be quite some time until full scale AI is available, so how far can we get with data in 2021 !?!
[edit] - I omitted to include a very important set of data points!! The above game vector data will only have currency of course when coupled with the live betfair market(s) data!!
IPT - Ideas spawned from the data experience
-
- Posts: 443
- Joined: Wed May 18, 2016 1:24 pm
Will be interesting to see how this thread develops - good first post, Jimibt.
My limited understanding is scanners like IPT are good, they are still not going to be near the data that the big syndicates use i.e. services like Opta.
Whilst unlikely to happen, would be interesting to have an expert match reader trading/betting alongside someone who uses data and see which one does better.
Perhaps the perfect storm would be someone who had expertise in both disciplines?
My limited understanding is scanners like IPT are good, they are still not going to be near the data that the big syndicates use i.e. services like Opta.
Whilst unlikely to happen, would be interesting to have an expert match reader trading/betting alongside someone who uses data and see which one does better.
Perhaps the perfect storm would be someone who had expertise in both disciplines?
-
- Posts: 3219
- Joined: Thu Oct 24, 2019 8:25 am
- Location: Newport
Good post Jim.
I might be being stupid but cant you just scan say the over goals markets and when you see layers are removing large volumes of money that this is a clear indication that something may about to be happening.
I might be being stupid but cant you just scan say the over goals markets and when you see layers are removing large volumes of money that this is a clear indication that something may about to be happening.
i forgot to mention in the OP. you make a good point. so yes, part of the decision making would be using live market data in conjunction with the real-time game vector data. i will amend the OP to reflect that thinking.Archery1969 wrote: ↑Tue Mar 16, 2021 9:59 amGood post Jim.
I might be being stupid but cant you just scan say the over goals markets and when you see layers are removing large volumes of money that this is a clear indication that something may about to be happening.
For anyone not aware of how the scanner looks I've setup a rule dedicated to Derek
Very simple, PI of 70 which indicates numerous SOT on & off over a certain period.
The underlying data is in JSON format tucked away in a PHP page, the site operates over drupal so it is possible theoretically to extract the data once you are logged in.
Very simple, PI of 70 which indicates numerous SOT on & off over a certain period.
The underlying data is in JSON format tucked away in a PHP page, the site operates over drupal so it is possible theoretically to extract the data once you are logged in.
You do not have the required permissions to view the files attached to this post.
as i said in the OP, i'm kinda trying to see if we can identify building blocks that in of themselves may not present much opportunity but when coupled with other vectors, produce strong signals that can be acted upon.
-
- Posts: 1074
- Joined: Fri Nov 20, 2015 9:38 am
AI tools are already available through :
https://cloud.google.com/ai-platform-un ... 1611054390
I've been messing about with it with some cricket data. Fairly straightforward tutorial... Basically get your data, specify the items you want to "learn" & the platform will select the most appropriate AI engine for your purposes..
It will run your data and give you a report on accuracy of model etc...
There is a $300 dollar credit available atm to although the processing costs seems to be relatively cheap..
It looks like you can link the engine up to live data feeds to give realtime decisions..
interesting - i've been using the ML.Net component in .net to do pretty much the same thing (using time based regression model). Still has a way to go as the actual sample data is of course king in all of this, so many revisions are required.sionascaig wrote: ↑Tue Mar 16, 2021 10:22 amAI tools are already available through :
https://cloud.google.com/ai-platform-un ... 1611054390
I've been messing about with it with some cricket data. Fairly straightforward tutorial... Basically get your data, specify the items you want to "learn" & the platform will select the most appropriate AI engine for your purposes..
It will run your data and give you a report on accuracy of model etc...
There is a $300 dollar credit available atm to although the processing costs seems to be relatively cheap..
It looks like you can link the engine up to live data feeds to give realtime decisions..
i'm still thinking about the traditional approach on this and in particular (becuase it is widely available), the IPT data and it's mix within the betfair odds etc.
-
- Posts: 1074
- Joined: Fri Nov 20, 2015 9:38 am
It looks ideal if you are a fairly competent trader: capture market data, capture trades, capture outcome ==> feed into model & should be able to build a baf in no time )
-
- Posts: 3219
- Joined: Thu Oct 24, 2019 8:25 am
- Location: Newport
This i am going to like this thread allot.
I think you know I'd be all over this Jim.
I'm using IPT PI data to drive some relatively simple strats already, alongside market data.
Does anyone know what formulas are used to derive PI1 and PI2 figures ? I know they are a mix of SOT's, pressure, corners etc over the previous x minutes
I'm currently pulling tons of API data for footy, which has quite a few more stats than those used by IPT. But the IPT stats have been tested for many years so assume they work well as is...
I'm using IPT PI data to drive some relatively simple strats already, alongside market data.
Does anyone know what formulas are used to derive PI1 and PI2 figures ? I know they are a mix of SOT's, pressure, corners etc over the previous x minutes
I'm currently pulling tons of API data for footy, which has quite a few more stats than those used by IPT. But the IPT stats have been tested for many years so assume they work well as is...