02-10-2012
, 03:32 AM
This is just the not-very-accurate error message after the parser bugs out on a player name with a space in it.
I should have a fix for this in the next update. You can work around this by deleting the space and making sure all the player names are just one word.
02-13-2012
, 04:43 PM
I'm not sure editing ranges for future rounds is going to happen anytime soon... For starters, it presents a real UI problem. A 6 handed game has 40 different ranges associated with it and over 300 possible future states... so, we're talking around 12000 different ranges. Even if you could edit future ranges, I dont think you'd actually want to.
02-14-2012
, 09:29 AM
Quote:
Thanks! I really appreciate it!
I'm not sure editing ranges for future rounds is going to happen anytime soon... For starters, it presents a real UI problem. A 6 handed game has 40 different ranges associated with it and over 300 possible future states... so, we're talking around 12000 different ranges. Even if you could edit future ranges, I dont think you'd actually want to.
I'm not sure editing ranges for future rounds is going to happen anytime soon... For starters, it presents a real UI problem. A 6 handed game has 40 different ranges associated with it and over 300 possible future states... so, we're talking around 12000 different ranges. Even if you could edit future ranges, I dont think you'd actually want to.
Maybe with simple opponent modeling, relative to equilibrium (different for pushing and calling).
For example we set opponent to "Loose pusher" which means his range is EQUILIBRIUM PUSH RANGE PERCENTAGE + (EQUILIBRIUM PUSH RANGE PERCENTAGE * 30%), in other words, if equilibrium is 20% then we add 30% of that range to that range which would be 20% + (20*0.3)% = 26%.
With this logic user could define his player types and how much range he wants to add/remove and UI wouldn't be a problem.
Thing is if there is a big stack fish who should push 100% of his range is very passive your future hands only calculate scenarios when big stack pushes (cause the probability is 100%) and it's ignoring real situations...
I know similar question was asked before, but how deep does the algorithm go when playing for example 3handed, I remember you said that deepness is a changing variable depending on something... Can you elaborate this mechanism?
02-14-2012
, 07:04 PM
Yeah, I think that some kind of opponent modelling is probably going to be the best approach to trying to "fine tune" the simulation of future rounds. And, I think that implying uncertainty about an opponents ranges is probably the best approach.
I dont even think that there would be much benefit to separating out pushing/calling ranges in such a system if the uncertainty is considered during the equilibrium calculation phases. You'd end up with whats called a "trembling hand" equilibrium... basically a subset of Nash that considers that some opponents might make mistakes.
Since version 1.0, the depth of the future simulation is always exactly 1 (so, 1 simulated future round of play... I consider the immediate hand to have a depth of 0) . During the beta, I had experimented with different mechanisms that would dynamically increase the depth of the game tree... for the whole tree if there were fewer players (say, 3), or only just a subset of the tree if there were more players.
Why limited to 1 future round now? Mostly to try to create a less complicated "baseline user experience". In most cases, the differences to the results from a deeper simulation are very very slight... and therefore rarely outweigh the often huge increase in processing time.
It is a certainty that I will put in the option to increase simulation depth at some point. But before I do, I want to be able to 1) quantify the benefit and 2) offset the performance hit. Since I'm currently working on my game simulator and the GPU implementation, you shouldnt have to wait *that* long...
I dont even think that there would be much benefit to separating out pushing/calling ranges in such a system if the uncertainty is considered during the equilibrium calculation phases. You'd end up with whats called a "trembling hand" equilibrium... basically a subset of Nash that considers that some opponents might make mistakes.
Since version 1.0, the depth of the future simulation is always exactly 1 (so, 1 simulated future round of play... I consider the immediate hand to have a depth of 0) . During the beta, I had experimented with different mechanisms that would dynamically increase the depth of the game tree... for the whole tree if there were fewer players (say, 3), or only just a subset of the tree if there were more players.
Why limited to 1 future round now? Mostly to try to create a less complicated "baseline user experience". In most cases, the differences to the results from a deeper simulation are very very slight... and therefore rarely outweigh the often huge increase in processing time.
It is a certainty that I will put in the option to increase simulation depth at some point. But before I do, I want to be able to 1) quantify the benefit and 2) offset the performance hit. Since I'm currently working on my game simulator and the GPU implementation, you shouldnt have to wait *that* long...
02-23-2012
, 01:42 PM
02-23-2012
, 11:41 PM
fluorescenthippo,
I guess the only thing I can say about the equilibrium ranges is, well... the math is the math.
Remember that while the equilibrium ranges are calculated to be unexploitable, they are not guaranteed to be the optimal strategy for a specific situation where your opponents might deviate from the equilibrium.
If you have specific knowledge about an opponents range, you certainly need to apply this information in order to get the real "answer".
Its a kind of "apples to oranges" thing to compare equilibrium strategies to what you believe to be a correct strategy based on your knowledge of actual player tendencies. For this reason, if you think that 10% is a better calling range than 20% in your example, then that may be true, but I would have to argue that it is more likely a coincidence and does not represent some kind of algorithmic advantage of an ICM-based Nash approximation over the Predictive Simulation in SnG Solver.
You say that no one would be calling A6o in the BB, but if UTG is pushing 38%, then they should. Likewise, if UTG if only pushing 28%, then SnG Solver agrees that the BB should fold A6o instead...
Spots like this are still going to come down to how well you can put your opponent on a range, and for this SnG Solver can try to put you in the ballpark, but you still have to do the rest of the work yourself.
And thanks for the feedback!
I guess the only thing I can say about the equilibrium ranges is, well... the math is the math.
Remember that while the equilibrium ranges are calculated to be unexploitable, they are not guaranteed to be the optimal strategy for a specific situation where your opponents might deviate from the equilibrium.
If you have specific knowledge about an opponents range, you certainly need to apply this information in order to get the real "answer".
Its a kind of "apples to oranges" thing to compare equilibrium strategies to what you believe to be a correct strategy based on your knowledge of actual player tendencies. For this reason, if you think that 10% is a better calling range than 20% in your example, then that may be true, but I would have to argue that it is more likely a coincidence and does not represent some kind of algorithmic advantage of an ICM-based Nash approximation over the Predictive Simulation in SnG Solver.
You say that no one would be calling A6o in the BB, but if UTG is pushing 38%, then they should. Likewise, if UTG if only pushing 28%, then SnG Solver agrees that the BB should fold A6o instead...
Spots like this are still going to come down to how well you can put your opponent on a range, and for this SnG Solver can try to put you in the ballpark, but you still have to do the rest of the work yourself.
And thanks for the feedback!
02-23-2012
, 11:55 PM
02-24-2012
, 12:18 AM