win rate confidence

bobman0330 · #1 01-10-2005, 05:31 PM

Would it substantially increase the accuracy of win rate if stat tracker or whoever estimated it by calculating a weighted (by hand frequency) average of individual hand win rates instead of simply aggregating all hands?

Piz0wn0reD!!!!!! · #2 01-10-2005, 07:34 PM

No.

Grisgra · #3 01-10-2005, 07:55 PM

Yes.

Piz0wn0reD!!!!!! · #4 01-10-2005, 08:08 PM

[ QUOTE ]
Yes.

[/ QUOTE ]

No. Specific hands take even longer to become accurate. I am just now shoing a profit w/ TT and 44 after 450 hands each.

pzhon · #5 01-10-2005, 10:51 PM

[ QUOTE ]
Specific hands take even longer to become accurate. I am just now shoing a profit w/ TT and 44 after 450 hands each.

[/ QUOTE ]
That doesn't mean it isn't more accurate to group the results by hand, then correct the hand frequencies.

You can decompose your luck into two pieces, first whether you are getting good starting hands, and second whether you are winning with each starting hand. Fixing the proportion of each starting hand is an attempt to eliminate the first form of luck, ideally leaving the second form plus your skill.

Piz0wn0reD!!!!!! · #6 01-10-2005, 10:58 PM

Perhapse it would be a tad more accurate. However that seems like a lot more trouble to go to for a small difference.

bobman0330 · #7 01-10-2005, 11:26 PM

Not if you're a computer program [img]/images/graemlins/tongue.gif[/img]

pzhon · #8 01-11-2005, 02:09 AM

I think it may be more accurate, but it might not be. The effect may be to eliminate the luck from which cards you are dealt, but to magnify the subsequent luck from the hands you are dealt infrequently. A priori, I don't see which effect dominates.

Something that you may have to resolve is how to handle hands with no representatives. For example, what value should be averaged in for A6s if after the first few thousand hands, you have never been dealt A6s? This becomes a more serious problem if you try to filter the hands not only by type, but also by position.