Jump to content

Neural Networks?


Recommended Posts

Re: Neural Networks?

Hello on this site there are predictions with neural networks http://www.prosoccer.gr/
Tried combining Prosoccer picks with Betvirus where both sites had a team at 60% or better but had to stop after 50 (real money) bets as was losing too much! Not sure if it might have been profitable after a longer run of bets or if I had used only selections from one site but I didn't track any more results.
Link to comment
Share on other sites

Re: Neural Networks?

That is a fantastic site. But it is difficult to work out whether the tips or profitable or not. Which suggests to me that they arent.
To be fair, as they provide tips for all matches played it is no chance in hell that it can be profitable. But I agree that sites like that are mostly pointless unless they lead running accounts on the performance of their tips, and anyone could put up a betting site with a lot of numbers and predictions.
Link to comment
Share on other sites

Re: Neural Networks?

Have a go with Easynn. It is free to trial and is not hard to get the hang of. I have not tried it with Soccer but will be only as good as the info. you put in.
EasyNN was one of the neural networks I tried but in the end I bought BrainMaker Pro, and despite many hours I have not been able to get it to work as I expected. There are so many design decisions to me made when creating your model. What Input/output data you want, What precision to use, how much data, how many neurons, what descriptive data to use,whether to use numerical changes or differences etc. etc. I have found the smallest change to your model results in a different output set and even when you run the same model with the same data you will even get a different output data set. When I go to NN forums and ask it seems that most people are using them for academic purposes and not practical purposes. I have spent a lot of time on NNs and seem to be getting nowhere.:wall
Link to comment
Share on other sites

Re: Neural Networks? A lot of stuff that you might have read regarding Neural Networks used for betting purposes have had a poor testing methodology. A proper testing method is to have say 10 seasons worth of data and use the first 8 for training and then use the last 2 for testing. However what you will find is that many of the websites offering NNs simply trained the NN on the whole 10 seasons and then published their prediction rate after say a 100 passes of the dataset as their achieved accuracy. This is simply nonsense as an NN will achieve very low training error if given enough passes through of the data, in fact I think it can theoretically achieve 0% training error which will lead to massive overfitting. Nonetheless, the site selling the NN call this their prediction accuracy without actual testing it on an unseen dataset and then just make easy money from people buying it. Also a major pitfall of NNs is that they are susceptible to the way data was received as an input. Consider three teams, A, B and C. If the games were A > B, B > C the probabilities the NN would give for A against C will not match those if the order of games received was B > C, A > B. This aspect of a NN has seen it basically been sidelined by the Machine Learning community in recent years. Just to clarify, I come from a Machine Learning background so am quite au fait with Neural Networks, Bayesian Networks, Kernel Regression and their pitfalls and successes. Frankly I believe a Bayesian Network approach is probably the best method to modeling football teams.

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...