01-24-2012
, 04:24 PM
Mirakulix,
Glad to hear you're up and running
The ranges (with PS disabled) are going to be a little different than holdemresources for couple reasons. Differences in handrankings is going to be one of them. Also differences in how we get the equilibrium to converge...
To be frank, generating a Nash equilibrium (or approximation thereof) for more than 2 players is dodgy business. For starters, there is no guarantee that there is only one solution... and they can be very numerically unstable... So, in order to guarantee that some kind of answer emerges at all in a timely manner you've got to pull out some "dirty tricks". And since I dont really know the details of what holdemresources is doing in this regard, its probably a bit different than what I'm doing.
Yes, any upgrades for the foreseeable future will be free. And I've got a lot of good stuff planned for that future.
I got your HHs... thanks. I dont really know if/how HEM might mangle the HHs after processing them... so I guess if you could send me a few before they go through HEM that would be good.
Thanks for the feedback!
Glad to hear you're up and running
The ranges (with PS disabled) are going to be a little different than holdemresources for couple reasons. Differences in handrankings is going to be one of them. Also differences in how we get the equilibrium to converge...
To be frank, generating a Nash equilibrium (or approximation thereof) for more than 2 players is dodgy business. For starters, there is no guarantee that there is only one solution... and they can be very numerically unstable... So, in order to guarantee that some kind of answer emerges at all in a timely manner you've got to pull out some "dirty tricks". And since I dont really know the details of what holdemresources is doing in this regard, its probably a bit different than what I'm doing.
Yes, any upgrades for the foreseeable future will be free. And I've got a lot of good stuff planned for that future.
I got your HHs... thanks. I dont really know if/how HEM might mangle the HHs after processing them... so I guess if you could send me a few before they go through HEM that would be good.
Thanks for the feedback!
01-24-2012
, 06:58 PM
Hi jason
You said that ICM is dead. Analising the results with the sngwiz and sngsolver i cant see to much diference.
What can you say to me and other people who are unsure about what programe to use cause we dont know each results are more reliable.
PS: send me the press release.
You said that ICM is dead. Analising the results with the sngwiz and sngsolver i cant see to much diference.
What can you say to me and other people who are unsure about what programe to use cause we dont know each results are more reliable.
PS: send me the press release.
01-25-2012
, 05:03 AM
yukoncpa,
You raise an important issue... I'll try to get in some kind of alternate/custom palette mechanism soon.
Pagasses,
In many situations, the results from SnG Solver and ICM-based programs will be nearly the same... these will be spots where the stacks are relatively deep and there is less of a "bubble effect" going on. But things can change quickly as the stacks get shallower and the blinds approach...
At the risk of making things more confusing, let me make an analogy to physics...
Modern physics describes a few fundamental forces in nature... gravity, electro-magnetism, and the nuclear forces. When considering the total force between two objects, all of the forces are present, but depending on the scale of the objects, the different forces become more or less important.
At the scale of planets and stars, gravity is by far the most dominant and the other forces are impossible to detect. This is like when the stacks are deep in a tournament. When the stack/blind ratio is very large, having to pay the blinds doesnt really matter and stack size is the only thing we need to calculate equity.
But when we consider a smaller scale... like the force of the magnet stuck to my refrigerator door, or the force that binds the molecules of the refrigerator itself, then the electro-magnetic force dominates over gravity. The gravity is still there, but it takes a back seat to the contribution of EM at this smaller scale. This is like the when the stacks get short and having to take the blinds means the loss of fold equity.
Finally, within the atom itself, the nuclear forces that keep the atomic nuclei from flying apart are far stronger than either EM or gravity at such a small scale. Again, the other forces are still there, but nowhere near as important. This is like when the stacks are smaller still and a micro stack UTG is at risk of busting and the "bubble effect" takes over.
In this analogy, ICM is like a tool that can only measure gravity. Its fine for measuring the force between planets... but terrible for measuring the forces of smaller objects. SnG Solver and its Predictive Simulation algorithm is sensitive to all of the forces... you can get accurate measurements at any scale.
Anyways, I hope that made some kind of sense. I've clearly been sitting at my computer too long and I'm starting to lose my mind...
But another thing to consider is the "edge" adjustment in SNGWiz. The fact that the "edge" adjustment exists at all is proof that there is something wrong with ICM based calculations. Any situation where you need to add in some kind of "edge" correction in order to make the results seem acceptable is a spot where ICM has failed. SnG Solver needs no such corrections to its results.
Also, dont forget that SnG Solver has some other things going for it besides Predictive Simulation... Like automatic calculation of approximate Nash strategies for all opponents means never forgetting to have to set opponent ranges. How many times have you read a thread in STT Strategy where somebody posts a sngwiz analysis thats ridiculous because they forgot to adjust all the default ranges?
Plus there's cool stuff like the EV-RMSD graph which, I must say, has not been getting enough love... but I'm sure you guys will come around eventually.
After the next update is out, I'll be working on some experiments to quantify just how much better SnG Solver is over the competition. I expect to be able to make a statement like: "If you use some other software to develop your strategies, you are being exploited by x% by someone using SnG Solver!"
Okay, tl;dr... right? I totally took your bait, Pagasses
You raise an important issue... I'll try to get in some kind of alternate/custom palette mechanism soon.
Pagasses,
In many situations, the results from SnG Solver and ICM-based programs will be nearly the same... these will be spots where the stacks are relatively deep and there is less of a "bubble effect" going on. But things can change quickly as the stacks get shallower and the blinds approach...
At the risk of making things more confusing, let me make an analogy to physics...
Modern physics describes a few fundamental forces in nature... gravity, electro-magnetism, and the nuclear forces. When considering the total force between two objects, all of the forces are present, but depending on the scale of the objects, the different forces become more or less important.
At the scale of planets and stars, gravity is by far the most dominant and the other forces are impossible to detect. This is like when the stacks are deep in a tournament. When the stack/blind ratio is very large, having to pay the blinds doesnt really matter and stack size is the only thing we need to calculate equity.
But when we consider a smaller scale... like the force of the magnet stuck to my refrigerator door, or the force that binds the molecules of the refrigerator itself, then the electro-magnetic force dominates over gravity. The gravity is still there, but it takes a back seat to the contribution of EM at this smaller scale. This is like the when the stacks get short and having to take the blinds means the loss of fold equity.
Finally, within the atom itself, the nuclear forces that keep the atomic nuclei from flying apart are far stronger than either EM or gravity at such a small scale. Again, the other forces are still there, but nowhere near as important. This is like when the stacks are smaller still and a micro stack UTG is at risk of busting and the "bubble effect" takes over.
In this analogy, ICM is like a tool that can only measure gravity. Its fine for measuring the force between planets... but terrible for measuring the forces of smaller objects. SnG Solver and its Predictive Simulation algorithm is sensitive to all of the forces... you can get accurate measurements at any scale.
Anyways, I hope that made some kind of sense. I've clearly been sitting at my computer too long and I'm starting to lose my mind...
But another thing to consider is the "edge" adjustment in SNGWiz. The fact that the "edge" adjustment exists at all is proof that there is something wrong with ICM based calculations. Any situation where you need to add in some kind of "edge" correction in order to make the results seem acceptable is a spot where ICM has failed. SnG Solver needs no such corrections to its results.
Also, dont forget that SnG Solver has some other things going for it besides Predictive Simulation... Like automatic calculation of approximate Nash strategies for all opponents means never forgetting to have to set opponent ranges. How many times have you read a thread in STT Strategy where somebody posts a sngwiz analysis thats ridiculous because they forgot to adjust all the default ranges?
Plus there's cool stuff like the EV-RMSD graph which, I must say, has not been getting enough love... but I'm sure you guys will come around eventually.
After the next update is out, I'll be working on some experiments to quantify just how much better SnG Solver is over the competition. I expect to be able to make a statement like: "If you use some other software to develop your strategies, you are being exploited by x% by someone using SnG Solver!"
Okay, tl;dr... right? I totally took your bait, Pagasses
01-25-2012
, 05:14 AM
Quote:
The Hero's range as shown in the Table view is the sum of all the hands that are +EV vs the ranges of the opponents. This is often slightly different than the calculated equilibrium range for the same position. This is because the equilibrium ranges (which are then used as the default opponent ranges) are restricted to a linear handranking to help speed up the calculations.
E.g. I solved for the good old Full Tilt Super Turbo starting scenario. (Each player starts with 300c, Blinds are 15/30)
Equilibrium says I should push 11.9%. Including thrashy hands like , A9o and 44, but no suited Aces.
The table view got rid of these hands and pushes hands that play better against these tight calling ranges. The problem is that it pushes now 15.1% including hands like JQo, KTo, K9s. I feel these hands are generally bad to push in the first place but more important it's definitely not unexploitable and once regs see me pushing that wide from UTG they will adjust and I get completely owned every time I push them.
So basically I'd like to see the 11.9% that are best and close to unexploitable be pushed from UTG.
BTW holdemressources and holdemressources.net push 8.6% / 8.8% from UTG. SnG solver without PS 10.3%.
Obviously the real equilibrium isn't found yet but it's hard for us to see which tool has the better educated guesses. I'd really like to see a match SnG Solver versus holdemressources
01-25-2012
, 05:26 AM
Quote:
Bump!
How is it going?
One big idea, maybe mentioned before...
I don't know how much information Sng solver needs to remember when hand is solved (cause i see that everything we do afterwards is fast so probably already computed) but if it isn't too much, idea is to make some kind of a buffer that remembers everything about hand. For example we solve and analyze hand "A", then we solve hand "B" and then we want to go back to hand "A" because we missed something. Now solver doesn't solve the hand cause solution is already in buffer and it's instantly shown.
If that's possible next step would be to make a button that allows us to solve multiple hands, and with that you get rid of most annoying thing in sng solver which is waiting, it's always between 5 sec to 1 minute and this time is too short to do anything else, so you can only watch progress bar slowly turning green.
The key for hands would be stack sizes, blind level and payout structure.
If all that matches a hand in buffer, copy information from buffer, else solve hand...
How is it going?
One big idea, maybe mentioned before...
I don't know how much information Sng solver needs to remember when hand is solved (cause i see that everything we do afterwards is fast so probably already computed) but if it isn't too much, idea is to make some kind of a buffer that remembers everything about hand. For example we solve and analyze hand "A", then we solve hand "B" and then we want to go back to hand "A" because we missed something. Now solver doesn't solve the hand cause solution is already in buffer and it's instantly shown.
If that's possible next step would be to make a button that allows us to solve multiple hands, and with that you get rid of most annoying thing in sng solver which is waiting, it's always between 5 sec to 1 minute and this time is too short to do anything else, so you can only watch progress bar slowly turning green.
The key for hands would be stack sizes, blind level and payout structure.
If all that matches a hand in buffer, copy information from buffer, else solve hand...
01-30-2012
, 03:57 AM
Quote:
... once regs see me pushing that wide from UTG they will adjust and I get completely owned every time I push them.
Quote:
....it's hard for us to see which tool has the better educated guesses.
You sure can do this... its going to use a ton of memory though. But it does raise a good point... It should be technically possible for multiple instances of the program to share the 700MB of pre-calced data. This would dramatically reduce memory usage per instance and make "multitabling" SnG Solver much more practical. I shall put this on my to-do list.
01-31-2012
, 02:58 PM
Quote:
So you basically calculate an equilibrium with restricted ranges, let hero adjust with unrestricted ranges and give villains no chance to counteradjust to the new range. So we basically have the restricted equilibrium range and an exploitive range in the table view? I see that it's to much work to give all players unrestricted ranges and I like the idea of giving hero unrestricted ranges (One of the biggest flaws of non beta holdemressources IMO) but there should be at least one point where you can see an unrestricted range that has the exact same percantages as the restricted equilibrium had. Otherwise you will often get 2 different ranges that both contain hands that shouldn't be in there.
E.g. I solved for the good old Full Tilt Super Turbo starting scenario. (Each player starts with 300c, Blinds are 15/30)
Equilibrium says I should push 11.9%. Including thrashy hands like , A9o and 44, but no suited Aces.
The table view got rid of these hands and pushes hands that play better against these tight calling ranges. The problem is that it pushes now 15.1% including hands like JQo, KTo, K9s. I feel these hands are generally bad to push in the first place but more important it's definitely not unexploitable and once regs see me pushing that wide from UTG they will adjust and I get completely owned every time I push them.
So basically I'd like to see the 11.9% that are best and close to unexploitable be pushed from UTG.BTW holdemressources and holdemressources.net push 8.6% / 8.8% from UTG. SnG solver without PS 10.3%.
Obviously the real equilibrium isn't found yet but it's hard for us to see which tool has the better educated guesses. I'd really like to see a match SnG Solver versus holdemressources
E.g. I solved for the good old Full Tilt Super Turbo starting scenario. (Each player starts with 300c, Blinds are 15/30)
Equilibrium says I should push 11.9%. Including thrashy hands like , A9o and 44, but no suited Aces.
The table view got rid of these hands and pushes hands that play better against these tight calling ranges. The problem is that it pushes now 15.1% including hands like JQo, KTo, K9s. I feel these hands are generally bad to push in the first place but more important it's definitely not unexploitable and once regs see me pushing that wide from UTG they will adjust and I get completely owned every time I push them.
So basically I'd like to see the 11.9% that are best and close to unexploitable be pushed from UTG.BTW holdemressources and holdemressources.net push 8.6% / 8.8% from UTG. SnG solver without PS 10.3%.
Obviously the real equilibrium isn't found yet but it's hard for us to see which tool has the better educated guesses. I'd really like to see a match SnG Solver versus holdemressources
Maybe I'm confused, but I thought that with multiple opponents in a game, being game theory optimal and being unexploitable are not synonymous. For example, in the beauty contest game that Phil Newall talks about in the, Intelligent Poker Player, the game theory optimal solution is zero ( whether heads up or multi player). When playing heads up, you will always tie or win when selecting zero, but when selecting the correct answer ( or game theory optimal answer ), zero in this case, in the multiplayer game, you will almost assuredly lose since your opponents aren't "playing along". It's my understanding that game theory optimal solutions in multiplayer games will guide you in the right direction but can never assure you of a situation that will always result in a tie or a win.
Am I wrong here? I do get easily confused
Last edited by yukoncpa; 01-31-2012 at 03:14 PM.
02-03-2012
, 12:53 AM
Okay.. the "update" installer packages now include the correct DLLs. The "full" installer takes a while to build and wont be up for a few more minutes.
Paul, can you let me know if this fixes your problem? (btw Paul... wow you're fast you already d/l'd, installed, and posted your problem before I even got to post my "1.0.3 is here" annoucement. heh.)
Paul, can you let me know if this fixes your problem? (btw Paul... wow you're fast
you already d/l'd, installed, and posted your problem before I even got to post my "1.0.3 is here" annoucement. heh.)
02-03-2012
, 12:32 PM
02-09-2012
, 05:58 AM
Okay gang, version 1.0.4 is ready for download.
Support for Ongame hand histories is now in, and I fixed a problem with reading some HHs from PokerStars.
I didnt really go over what went into the 1.0.3 update, so I'll also mention a few things of note from that version that you might not have noticed...
- Hand history import! At last. Support is still relatively basic, but will eventually become more sophisticated. You can either paste in a single hand, or load an entire multi-hand file and pick the one you're interested in. Currently supported sites are: PokerStars, Everleaf, Cassava/888/Pacific, and now Ongame. Definitely send me the HHs for any other site you'd like to see supported.
- Customizable colors. From the Tools->Options menu you can now choose an alternate color palette or create your own. I find myself using the "Contrast 1" palette a lot... it's very soothing.
- Better performance. I spent a lot of time re-working the equilibrium calculation to make it more efficient... depending on the exact game setup, you might see as much as a 40% reduction in solve times. Also, if you have a CPU that supports SSE4.1 instructions, SnG Solver will now take advantage of that fact... and that's worth another 5% or so speedup.
I mentioned this in another thread (in STT Strategy), but I'll mention it here too...
I've finished writing a simulator that plays complete STTs from start to finish... a perfect testbed to compare different strategy algorithms and equity models. I'm currently running a matchup between ICM-based Nash strategies and Nash strategies based on the Predictive Simulation model from SnG Solver. Im using a 6max format, with a "Super Turbo" type structure. 10BB starting stacks and 65/35 payouts. 3 players use PSM-Nash and 3 use ICM-Nash. It takes about 1 minute to simulate one complete STT, and I intend to play about 10000 simulated games... so it's going to take a while to finish
But even after 1200 games, I think the results are pretty interesting...
So, for anyone who was wondering just how much better than ICM you can get... the answer just might be: good enough to turn an ICM-based Nash strategy into a -4% RoI loser...
Support for Ongame hand histories is now in, and I fixed a problem with reading some HHs from PokerStars.
I didnt really go over what went into the 1.0.3 update, so I'll also mention a few things of note from that version that you might not have noticed...
- Hand history import! At last.
Support is still relatively basic, but will eventually become more sophisticated. You can either paste in a single hand, or load an entire multi-hand file and pick the one you're interested in. Currently supported sites are: PokerStars, Everleaf, Cassava/888/Pacific, and now Ongame. Definitely send me the HHs for any other site you'd like to see supported.- Customizable colors. From the Tools->Options menu you can now choose an alternate color palette or create your own. I find myself using the "Contrast 1" palette a lot... it's very soothing.
- Better performance. I spent a lot of time re-working the equilibrium calculation to make it more efficient... depending on the exact game setup, you might see as much as a 40% reduction in solve times. Also, if you have a CPU that supports SSE4.1 instructions, SnG Solver will now take advantage of that fact... and that's worth another 5% or so speedup.
I mentioned this in another thread (in STT Strategy), but I'll mention it here too...
I've finished writing a simulator that plays complete STTs from start to finish... a perfect testbed to compare different strategy algorithms and equity models. I'm currently running a matchup between ICM-based Nash strategies and Nash strategies based on the Predictive Simulation model from SnG Solver. Im using a 6max format, with a "Super Turbo" type structure. 10BB starting stacks and 65/35 payouts. 3 players use PSM-Nash and 3 use ICM-Nash. It takes about 1 minute to simulate one complete STT, and I intend to play about 10000 simulated games... so it's going to take a while to finish
But even after 1200 games, I think the results are pretty interesting...
So, for anyone who was wondering just how much better than ICM you can get... the answer just might be: good enough to turn an ICM-based Nash strategy into a -4% RoI loser...