#11
|
|||
|
|||
Re: How many hands do you need to determine true win rate?
It depends on your Standard Deviation (which depends upon your style of play, the games you play in etc).
|
#12
|
|||
|
|||
Re: How many hands do you need to determine true win rate?
This is a good benchmark that I stole from MrWookie from Micros
[ QUOTE ] My personal metric is, for play at a particular level, to know to within two standard errors (about 98% confidence) that I am a winning player. That is, that my win rate is twice or more as big as the uncertainty of my win rate. To compute the uncertainty of your win rate, take your standard deviation per 100 hands, usually about 15 BB/100, and divide it by the square root of the number of hands you’ve played divided by 100 (the number of 100 hand blocks you’ve played). Playing 20,000 hands with this standard deviation will yield an uncertainty in your win rate of 1.06 BB/100. Thus, you’d need a win rate of 2.12 BB/100 to know with 98% confidence that you were a winning player. Depending on your personal level of boldness or paranoia, you may be satisfied with 84% confidence (uncertainty = win rate) or 99.9% confidence (uncertainty = win rate / 3). Of note is that 20,000 hands at a win rate of 2.12 BB/100 will net you 424 BB, which, combined with the 300 BB you started with, gives you a bankroll sufficient to play at twice the current limit. An interesting side effect of this metric is that players who are truly crushing a particular level will advance more quickly, since they will need fewer hands to reduce their uncertainty to half (or whatever) their win rate, while players winning less are encouraged to stick around longer, hopefully learning new lessons that bring their win rate up along the way. [/ QUOTE ] |
#13
|
|||
|
|||
Re: How many hands do you need to determine true win rate?
[ QUOTE ]
299,792,458 [/ QUOTE ] Geek. What's worse is that it hurts me to see it without units. |
#14
|
|||
|
|||
Re: How many hands do you need to determine true win rate?
[ QUOTE ]
[ QUOTE ] 299,792,458 [/ QUOTE ] Geek. What's worse is that it hurts me to see it without units. [/ QUOTE ] The units is number of hands. |
#15
|
|||
|
|||
Re: How many hands do you need to determine true win rate?
[ QUOTE ]
[ QUOTE ] [ QUOTE ] 299,792,458 [/ QUOTE ] Geek. What's worse is that it hurts me to see it without units. [/ QUOTE ] The units is number of hands. [/ QUOTE ] Is they? |
#16
|
|||
|
|||
Re: How many hands do you need to determine true win rate?
I hope I'm around to see the mass suicide that ensues when you all find out these t tests and stochastic models you're in love with are complete bull [censored] in this circumstance.
(This reply isn't really directed as snowbank in particular) |
#17
|
|||
|
|||
Re: How many hands do you need to determine true win rate?
[ QUOTE ]
[ QUOTE ] [ QUOTE ] [ QUOTE ] 299,792,458 [/ QUOTE ] Geek. What's worse is that it hurts me to see it without units. [/ QUOTE ] The units is number of hands. [/ QUOTE ] Is they? [/ QUOTE ] Of course. He asked for a number of hands. omg. |
#18
|
|||
|
|||
Re: How many hands do you need to determine true win rate?
here I'll show you how to do it. your sample's SD is (Sample SD in BB/100)/Sqrt(Hands/100). you want 3 SD's since that represents the statistically significant region.
you can use excel to find it for you. in A1 put your SD in BB/100. 18 is pretty typical in A2 put an arbitrary number. this will represent the # of hands in A3 put =A1/SQRT(A2/100)*3 then go to tools-goal seek: Set Cell: A3 To Value: .25 By changing cell: A2 your answer will be in A2. I'm getting about 4.7 million |
#19
|
|||
|
|||
Re: How many hands do you need to determine true win rate?
[ QUOTE ]
I hope I'm around to see the mass suicide that ensues when you all find out these t tests and stochastic models you're in love with are complete bull [censored] in this circumstance. (This reply isn't really directed as snowbank in particular) [/ QUOTE ] if you're talking about Justin's findings, he hasn't found anything yet really. it's skewed. we don't really know why that is. it should still be more or less accurate. |
|
|