Probability output

GarettGarett Member Posts: 10 Contributor II
edited November 2018 in Help
Hello,
This is my introductory post so I wanted to say hello. :)

Also, rapidminer is a wonderful application and thanks to the developers.

I am planning on using rapidminer to predict the probability of a person folding in different situations in online poker, and I wanted to ask for the best architecture to return a value between 0 and 1, which is the predicted probability of someone folding their hand.

I believe neural nets and supervised learning would be neccesary using inputs from thousands of hand histories I have saved.  I can export certain player statistics to excel.  I think it would contain certain attributes like:

player1-amount commited to pot ----- player 2 amount commited to pot ---- betting round ---- texture of the board...


Then I want the label or the output value to be the corresponding probability that was learned from the hand histories.  This probability would be added to the probability of winning in the showdown based on many monte carlo showdowns to give me my value to raising.

Would neural nets be best for what I'm trying to do?
Tagged:

Answers

  • GarettGarett Member Posts: 10 Contributor II
    I think I posted this in the wrong forum  :o. Sorry.
  • GarettGarett Member Posts: 10 Contributor II
    Ok, did some research, looks like NaiveBayes would be my best bet per this article about probability estimation:
    http://www.cs.washington.edu/ai/nbe/nbe_icml.pdf
  • IngoRMIngoRM Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, Community Manager, RMResearcher, Member, University Professor Posts: 1,751 RM Founder
    Hi,

    nice application, and if you manage to personalize your predictions to specific opponents it should also become profitable  ;)

    Would neural nets be best for what I'm trying to do?
    Probably not. From my experience, a well tuned neural network can do well on small amounts of data but is usually not very robust and does not scale well up to large amounts of data. There is of course no silver bullet which always works fine, but I would definitely give data mining methods like NaiveBayes, the various SVM implementations and other regression schemes as well as more simple methods like Perceptron a go.

    Cheers,
    Ingo
Sign In or Register to comment.