[jukofyork]

[ QUOTE ]
[ QUOTE ]
[ QUOTE ]
Pokergirl z
Pre-flop played %
65

grego777
Pre-flop played %
66

Daurgman
Pre-flop played %
68

[/ QUOTE ]

It's very unlikely that these stats in particular came from the same player/bot. If the player's true stat is 65, then after 62K hands they will be within that ~.4 of the 65 95% of the time. It's about a one in ten million event that this player would have a 66 stat. It's virtually impossible for the stat to be 68 over the samples discussed. Of course, this type of analysis assumes things that may not be true such as static game conditions/inability of the bot to adjust. It also assumes that MH or FT didn't round 65.4 to 65 and 65.6 to 66. Even so, I felt it necessary to point out that appearances here maybe deceiving and that it's possible to draw the opposite conclusion from the data then is obvious from just looking at it.

[/ QUOTE ]

Leader,

It seems as if this new batch of bots learned from their old mistakes and are taking precautions. It's well known now that the sites (and many concerned players too) are making stat comparisons as a way of identifying potential bots.

It would be very easy to make the bot configurable so that stats don't converge exactly, even though the underlying logic is essentially the same. Think, for example, of making small adjustments to stealing and defending ranges that will have very little impact on your bottom line.

[/ QUOTE ]
I agree with this. Just looking at PokerAcademy's "Sparbot" it's obvious how this could be accomplished. Basically they created two different "Sparbots":

a) The first "Sparbot" has no preference for choosing a fold, call or raise action in any given state.

b) The second "Sparbot" uses the idea that humans tend to make the most mistakes when faced with aggressive play so if both a call and a raise action have the same EV, then it will always choose the raise, etc.

The two models play quite differently, yet both are still approximating the NE solution and I guess would have quite different stats. You can then combine the mixed F/C/R output (ie: probability triples) of the two systems using a weighted average (eg: 30% SparbotA + 70% SparbotB) to create all sorts of different "hybrid" bots.

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