I'm thinking of tightening up in these kinds of spots (pf) - Page 2

Dendrite · #11 08-14-2006, 09:49 PM

[ QUOTE ]

how many hands do you think you need in your db to have a good sample size for K9o vs a raise and a caller in the big blind? say you have 1000000 hand db, how skewed do you think this sample might be by having maniacs or tight passives in your game at the time??

[/ QUOTE ]

it doesn't have to be perfect bb/h. why not analyse KT K9 K8 QT Q9 Q8 in the BB 3way, and see what your bb/h is? if it's -.2, then it's clearly profitable. if it's -.8, then something's going wrong. if it's in the middle, then it's more murky. but so what? this kind of analysis can be useful. what do you use pokertracker for, tracking winnings only?

everytime i suggest using empirical analysis on this board people always say "LOL SAMPLE SIZE", but their theoretical analysis of profitbility in preflop situations is absolutely abysmal as well. we don't have any alternatives, so let's make do with what we've got.

Dendrite · #12 08-14-2006, 10:07 PM

I went to pokertracker. I filtered for the following:

I am in the hijack of a 4-6 handed game. Exactly 3 people saw the flop, no more no less. I wanted to see how my hands were performing in this exact situation.

"O, but Dendrite! Your conditions are too narrow! Your database isn't 3 million hands!"

(Pocket pairs shown - because we can easily compare this graph to what the values "should" be)

It's a little rough, but the information is there and useable for analysis. Note that each bar only represents 6 combos of hands. If they are grouped together, like I mentioned earlier - say into AA-QQ, JJ-99, etc, then the avg group performance becomes closer to reality and helps lessen the pain of having a small sample.

Leader · #13 08-14-2006, 10:10 PM

[ QUOTE ]
how are you gonna model what happens after the flop?

[/ QUOTE ]

We're going to look generally at the things that make up the biggest part of our ev post flop. Like when we make a pair and win, when we make a pair and lose, and when we hit nothing.

[ QUOTE ]
we know how often you flop a pair. now many BBs are you going to make when you flop a pair? there's no way to calculate it.

[/ QUOTE ]

You can estimate it though. I'm not saying you can make a rock hard statement about what to do in these marginal spots. What I'm saying is that this spot is marginal and how you decide what to do is to look at post flop. If you play most of the time you’re going to make nothing and fold. The question is are you happy when you hit and how hard do you have to hit so that you feel like the post flop bets are going in with you as the favorite rather then going in with you as the dog but with the pot laying you to much to fold.

[ QUOTE ]
if you wanna know if a hand is gonna be profitable, you have to be able to estimate how much money you will make/lose after the flop in these situations. and since it's impossible to do, we just have to look at empirical data. all these discussions always degenerate into relative arguments - "I'd rather call with XX than YY" - but never absolute arguments - "JJ will make me -.55 bb/h if I call, KK will make me -.45 bb/h if I call".

you can help alleviating sample size issues by grouping together similiar hands which will give you a larger sample.

[/ QUOTE ]

Ok let’s group all the hands together and see how we're doing. That's call BB/100 and it takes 400-500K to converge to even +- .5BB/100. You want to take a 2000th of that sample and draw conclusions about whether a hand that is likely between -.1BB and +.1BB? I'll take my theoretical abstractions and educated guesses.

Leader · #14 08-14-2006, 10:12 PM

Maybe you could elaborate on what that shows. I don’t get it.

Leader · #15 08-14-2006, 10:14 PM

[ QUOTE ]
we don't have any alternatives,

[/ QUOTE ]

Just because you don't understand the alternative doesn't mean it's nonexistent.

Dendrite · #16 08-14-2006, 10:16 PM

The reason BB/100 takes 500K to converge to +- .5BB/100 because there are so many different situations to play. Individual situations will converge much quicker than that.

If I give you AA on the button for every hand you're ever dealt in the future it's not going to take 500K hands for it to get +- .5BB/100. It will converge quicker than that. Same with K9 in the BB against a raiser and a coldcaller with roughly the same ranges. And we don't need to even need to know a .5BB/100, that works out to be 0.005 BB / hand. I don't care about that kind of resolution. Just knowing the EV of the hand within an error +/- 0.15 BB / hand (which works out to be +/- 15 BB / 100) is enough for us to tell us if it's "clearly profitable" "pretty close" or "clearly unprofitable"

Leader · #17 08-14-2006, 10:23 PM

[ QUOTE ]
The reason BB/100 takes 500K to converge to +- .5BB/100 because there are so many different situations to play. Individual situations will converge much quicker than that.

[/ QUOTE ]

I agree, but they also come up much less frequently. And I don't think they converge anywhere near as fast as you think.

[ QUOTE ]
Just knowing the EV of the hand within an error +/- 0.15 BB / hand (which works out to be +/- 15 BB / 100) is enough for us to tell us if it's "clearly profitable" "pretty close" or "clearly unprofitable"

[/ QUOTE ]

Note that "pretty close" is the only one that matters here. Almost everyone’s strategy has the other two down. So it does need to be very accurate.

Dendrite · #18 08-14-2006, 10:35 PM

For example -

I filtered for:

Blind status: BB
VPIP: Put Money In (this means someone else raised)
Hands with between: 3 and 3 players seeing the flop
Hole cards: K9o K8o Q9o Q8o JTo J9o J8o

BB/Hand: (0.02)

This qualifies as ridiculously profitable.

The earlier graph is meant to show how even with similiar amounts of drilling down (Specific position, specific number of people seeing the flop), the results are pretty close to 'reality' anyway.

The grouping of similiar hands makes helps, too.

Dendrite · #19 08-14-2006, 10:42 PM

If someone runs that exact same query and gets in the neighborhood of (1.00) BB/hand I will be very suprised

Leader · #20 08-15-2006, 01:21 AM

[ QUOTE ]
If someone runs that exact same query and gets in the neighborhood of (1.00) BB/hand I will be very suprised

[/ QUOTE ]

LMAO That's exactly what I got when I ran this through 17K of July and August stats. Well I guess 2 hands isn't a big enough sample. At this rate, I'll figure out my win rate here in a few decades.