Ideas on automating table selection

[bziegler2]

[ QUOTE ]
wouldn't your own personal style need to be involved somewhere? I'd think a TAG would like different tables than LAGS would.

[/ QUOTE ]

Yes, that's right. You'd need a different scoring function for Table Score based on the player's own style.

[jukofyork]

[ QUOTE ]
My problem is I like to write poker software, but I just don't play that much. I only have about 5000 hands played. Something like this is going to need about 50k hands on 1 player, and it would be good to have multiple players with different styles to see if the formulas are consistent.

Does anyone want to donate some data to get this effort off the ground?

[/ QUOTE ]
I'd be pretty interested to see how you get on and I'm happy to donate, but it might take a bit to get the files to you (I can upload them somewhere for you to dl if you want?).

[ QUOTE ]
I forgot to mention, I think I have an easy way to figure out some good weighting numbers. I could use the "Goal Seek" function in Excel to change the weights until there is a maximum correlation between the Table Score and the player's win rate at the table.

[/ QUOTE ]
Just read up on this and it sounds quite interesting (I hardly use excel, so never seen this before). Any reason to think this works better than just using least-squares minimization? Also, how does it get on with non-linear equations?

Juk [img]/images/graemlins/smile.gif[/img]

[WolfElder]

[ QUOTE ]
[ QUOTE ]
isnt this similar to that what the program poker table scan does?

[/ QUOTE ]

I don't know of this poker table. Can you describe it?

[/ QUOTE ]

its talked about at length in another thread here..

http://forumserver.twoplustwo.com/sh...age=1&vc=1

I just thought it was similar to what you guys were talking about.

[bziegler2]

Juk,

I got a bunch of data from one 2+2 user, about 130k hands of 100NL & 200NL.

Getting good results already. Using a basic "Fish Score", we can identify tables are better. For the top 10% of tables based on Fish Score, his winrate was double what it was overall. For the top 35% it was 20-50% higher. So as no surprise, good table selection definitely pays off.

Right now I'm working on optimizing the parameters for the fish score. It might be a least scores fit problem. It probably is since there is no interaction among the parameters.

My main problem is that there is a bunch of data, and I can't easily get it into Excel. So I'm just going to bite the bullet and write a quick program to load in the data and do the calcs. I'll let you know if the new fish scores can rate the tables much better than the default score (which was just using the predicted BB/100 as we discussed before).


Wolf,

Thanks for pointing out that other thread. Once we finish here, someone should let that author know what we came up with, so he could incorporate it into his program.

-Ben

[jukofyork]

[ QUOTE ]
Juk,

I got a bunch of data from one 2+2 user, about 130k hands of 100NL & 200NL.

Getting good results already. Using a basic "Fish Score", we can identify tables are better. For the top 10% of tables based on Fish Score, his winrate was double what it was overall. For the top 35% it was 20-50% higher. So as no surprise, good table selection definitely pays off.

Right now I'm working on optimizing the parameters for the fish score. It might be a least scores fit problem. It probably is since there is no interaction among the parameters.

My main problem is that there is a bunch of data, and I can't easily get it into Excel. So I'm just going to bite the bullet and write a quick program to load in the data and do the calcs. I'll let you know if the new fish scores can rate the tables much better than the default score (which was just using the predicted BB/100 as we discussed before).


Wolf,

Thanks for pointing out that other thread. Once we finish here, someone should let that author know what we came up with, so he could incorporate it into his program.

-Ben

[/ QUOTE ]
Can I ask what parameters you are using as import to the fish-score function? (VPIP, PFR, etc?).

Sounds very cool. I've got bogged down with other stuff and not had time to try improving on my LWS idea yet, but this looks hopeful.

Juk [img]/images/graemlins/smile.gif[/img]

[bziegler2]

Currently it uses these rules. They were taken from NL100 & NL200 data:

when VOL_PUT_IN < 21.44 and STEAL_PER between 19.66 and 28.25 and AGGRO_FAC >= 3.05 then 1.7
when VOL_PUT_IN < 21.44 and STEAL_PER between 15.59 and 19.66 then 1.4
when VOL_PUT_IN < 21.44 and STEAL_PER < 11.18 and AGGRO_FAC >= 2.72 then 1.1
when VOL_PUT_IN >= 21.44 and AGGRO_FAC >= 3.24 then 1.0
when VOL_PUT_IN < 21.44 and STEAL_PER between 11.18 and 15.59 then 0.6
when VOL_PUT_IN < 21.44 and STEAL_PER between 19.66 and 28.25 and AGGRO_FAC < 3.05 and SEE_SHOWDOWN_PER >= 31.86 then 0.5
when VOL_PUT_IN < 21.44 and STEAL_PER < 11.18 and AGGRO_FAC < 2.72 then 0.2
when AGGRO_FAC < 2.25 and VOL_PUT_IN between 21.44 and 31.04 and SEE_SHOWDOWN_PER >= 31.33 then 0.1
when VOL_PUT_IN >= 21.44 and AGGRO_FAC between 2.25 and 3.24 then 0.1
when VOL_PUT_IN < 21.44 and STEAL_PER between 19.66 and 28.25 and AGGRO_FAC < 3.05 and SEE_SHOWDOWN_PER < 31.86 then -0.3
when VOL_PUT_IN < 21.44 and STEAL_PER >= 28.25 then -0.4
when AGGRO_FAC < 2.25 and VOL_PUT_IN between 21.44 and 31.04 and SEE_SHOWDOWN_PER < 31.33 then -0.6
when AGGRO_FAC < 2.25 and VOL_PUT_IN >= 31.04 then -1.3