NL Bots on Full Tilt

[ShaneP]

[ QUOTE ]
[ QUOTE ]
[ QUOTE ]
[ QUOTE ]
[ QUOTE ]
- I argue that the reason all of the data is so similar is that he is playing on all of them. Think about this: Take 400k hands and split them up into four 100k chunks. Won't the data for each of these 100k hands be very similar to each other?

[/ QUOTE ]

But VPIP, which should obviously be identical across the board, is statistically different between the 4 accounts listed.

[/ QUOTE ]

How about this: At some time t0, he decided he wanted to adjust something in the strategy. He has four accounts with x1, x2, x3, anx x4 hands each. But say they're all different # of hands. The stats for each of those will be different based on how many hands he played at his strategy before t0 and the number of hands he plays with the new strategy after t0.

Example:

Set1: 110
Set2: 111100

Both have average of .67.

Change strategy to play all hands. Add 1 to each

Set1: 110 1
Set2: 111100 1

Average for 1: 0.75
Average for 2: 0.71

Same strategy, different averages, hmm. Maybe theres is a lesson here? [img]/images/graemlins/wink.gif[/img]

[/ QUOTE ]

A valid point.

However, two of the accounts have basically the same number of hands (105k hands and 112k hands), so we can assume they were datamined at the same time. How come their VPIPs are so significantly different then? (13.64% and 14.08%) That difference is over 4 SDs

[/ QUOTE ]

has it been confirmed that the true deviation is over 4SDs? earlier there were like 4 formulas people were trying to use. if true this fact needs much more attention as it has the greatest chance of clearing them

[/ QUOTE ]

I did a few quick calcs...not sure if this is 100% accurate, but I used the correct formula for SD of percentages (it is sqrt( p*(1-p)/n ). The null hypothesis was that they all came from the same distribution, with the mean being the weighted average of all four preflop players (the last line is the sum of all of them). First two columns are the VPiP and number of hands, the third is the individual percentage VPiP. The fourth is the SD using the number of hands of each individual player, and the last is the number of SD away from the mean (13.93% VPiP).

VPiP #Hands pct SD SD away
from mean
14376 | 105366 |0.136438699 | 0.001066697|-2.675179739
15840 | 112514 |0.14078248 | 0.001032258|1.443606827
11683| 82577 |0.141480073 | 0.00120493|1.815679571
5721 | 41414 | 0.138141691 | 0.001701445|-0.676257655

47620 |341871 |0.139292306

So only the first (one4thethumb I believe) is statistically different than 13.92, at the 95% level, but that's not really a result to hang one's hat on. The others are within 2 SD of the mean.

The greater than 4 SD different is adding the two SD from the first two players.

However, add to the SD above slight tweaks to code (assuming they're running bots) or human play that changes things slightly, and I can't reject the null hypothesis that these results weren't generated from the same process (the same mean).

Shane