*SnG Solver* - open beta
12-07-2011
, 07:18 PM
12-13-2011
, 03:19 PM
Amen
Quote:
OKAY 
I think if we look at some basic numbers, some pretty clear distinctions will emerge. Lets compare some equity estimations between SitNGo Wizard, holdemresources.net, and SnG Solver. Calculating stack equity is going to be the *core* of any kind of strategy analysis we eventually want to make.
Lets imagine a simple scenario... but one where the players are short stacked, short handed, and near the bubble. We've got 5 players with 500 chips each. Blinds are 100/50 with a 500-300-200 payout structure.
So first thing actually, lets see what a "standard" ICM calculator will say about this scenario:
Since the standard ICM model only takes stack sizes and payouts as inputs, it can only think that equal stacks must have the same equity. Everyone (hopefully) knows this does not match reality. We know that the 500 chips on the BTN is worth more than the 500 chips UTG. We may not be able to say how much more, be we know that it must be more because of the advantage of position. So thats problem #1 for the standard ICM model.
Problem #2 is that it doesnt make sense that the SB and BB have the same equity since they are forced to commit some of their chips into the pot whether they like it or not. Clearly, they should have at least some amount of reduced equity relative to the BTN since they have donated chips to the pot that are now vulnerable to attack.
Already, it should hopefully be clear that standard ICM all by itself has some serious issues.
Okay, let put our scenario into the calculator at holdemresources.net:
Okay, well there's clearly some improvement in representing the loss of equity from the stacks in the blinds. But right away, we see the values from the other positions dont match what we know to be true in reality.
According to this anaylsis the stack UTG is more valuable than the stack on the BTN. If you were sitting at the BTN in this tournament would you want to trade seats with the UTG player? Me neither.
SitNGo Wizard doesnt seem to have an "equity display" that just shows the value of each stack (unless I'm missing something... someone please correct me if I am). But you can get a "EQ Fold" for each stack. This is not exactly the same thing as stack equity, but these values should follow the same pattern as what a calculated total equity would be.
Here again, we've got some numbers that are not what we should expect. Like holdemresrouces.net, SitNGo Wizard is also trying to tell us that the stack UTG has an equity advantage over the rest of the table.
So now lets see what SnG Solver says.
Ah ha! At last, some numbers that make sense... 
The BTN has the most equity, followed by the CO. Next, even though the SB has to put up some "dead" equity, the fact that it has good position (both for attack potential on the BB and that it has a long way to go before the blinds) has made up for it. Finally, the BB is clearly the loser here... lots of "dead" equity, it is vulnerable to attack and the next hand it must put up even more of its stack in the SB. These are numbers that are consistent with good poker thinking.... and therefore the best chance to a good foundation to support later analysis.
Again, I am not trying to bash either holdemresources or SitNGo Wizard... they have both been a tremendous benefit to the poker community. But its also true that they both have some fundamental limitations that keep their analysis from being the best possible.
I think if we look at some basic numbers, some pretty clear distinctions will emerge. Lets compare some equity estimations between SitNGo Wizard, holdemresources.net, and SnG Solver. Calculating stack equity is going to be the *core* of any kind of strategy analysis we eventually want to make.
Lets imagine a simple scenario... but one where the players are short stacked, short handed, and near the bubble. We've got 5 players with 500 chips each. Blinds are 100/50 with a 500-300-200 payout structure.
So first thing actually, lets see what a "standard" ICM calculator will say about this scenario:
Code:
"Standard" ICM
Equity
UTG: 200.0
CO: 200.0
BTN: 200.0
SB: 200.0
BB: 200.0
Problem #2 is that it doesnt make sense that the SB and BB have the same equity since they are forced to commit some of their chips into the pot whether they like it or not. Clearly, they should have at least some amount of reduced equity relative to the BTN since they have donated chips to the pot that are now vulnerable to attack.
Already, it should hopefully be clear that standard ICM all by itself has some serious issues.
Okay, let put our scenario into the calculator at holdemresources.net:
Code:
holdemresources.net ICM Nash Calculator Results EQ post UTG: 209.4 CO: 209.4 BTN: 209.3 SB: 193.2 BB: 178.7
According to this anaylsis the stack UTG is more valuable than the stack on the BTN. If you were sitting at the BTN in this tournament would you want to trade seats with the UTG player? Me neither.
SitNGo Wizard doesnt seem to have an "equity display" that just shows the value of each stack (unless I'm missing something... someone please correct me if I am). But you can get a "EQ Fold" for each stack. This is not exactly the same thing as stack equity, but these values should follow the same pattern as what a calculated total equity would be.
Code:
SitNGo Wizard
"EQ Fold"
UTG: 216.0
CO: 209.8
BTN: 203.7
SB: 189.5
BB: 168.3
So now lets see what SnG Solver says.
Code:
SnG Solver
Equity
UTG: 189.1
CO: 216.4
BTN: 218.3
SB: 201.3
BB: 174.9
The BTN has the most equity, followed by the CO. Next, even though the SB has to put up some "dead" equity, the fact that it has good position (both for attack potential on the BB and that it has a long way to go before the blinds) has made up for it. Finally, the BB is clearly the loser here... lots of "dead" equity, it is vulnerable to attack and the next hand it must put up even more of its stack in the SB. These are numbers that are consistent with good poker thinking.... and therefore the best chance to a good foundation to support later analysis. Again, I am not trying to bash either holdemresources or SitNGo Wizard... they have both been a tremendous benefit to the poker community. But its also true that they both have some fundamental limitations that keep their analysis from being the best possible.
12-17-2011
, 01:30 AM
Hi !
I am wondering if you use some sort of Monte Carlo approach to solve each AICM situation over "millions of hands" ? If you do, it would be nice to allow the user some control over the calculation time (think pokerstove and/or a keyboard input). Currently, your AICM only gives a complete *perfect* evaluation. This is kind of lengthy for some computers to handle when a shorter *still more accurate than sngwiz* evaluation would be more than enough for many situations. Would your AICM lose a significant degree of accuracy if the computing time were cut into half ?
Also, any thoughts on a release date as I can't find that info anywhere ! Thanks !!! This all looks interesting
I am wondering if you use some sort of Monte Carlo approach to solve each AICM situation over "millions of hands" ? If you do, it would be nice to allow the user some control over the calculation time (think pokerstove and/or a keyboard input). Currently, your AICM only gives a complete *perfect* evaluation. This is kind of lengthy for some computers to handle when a shorter *still more accurate than sngwiz* evaluation would be more than enough for many situations. Would your AICM lose a significant degree of accuracy if the computing time were cut into half ?
Also, any thoughts on a release date as I can't find that info anywhere ! Thanks !!! This all looks interesting
12-17-2011
, 11:02 AM
Quote:
Hi !
I am wondering if you use some sort of Monte Carlo approach to solve each AICM situation over "millions of hands" ? If you do, it would be nice to allow the user some control over the calculation time (think pokerstove and/or a keyboard input). Currently, your AICM only gives a complete *perfect* evaluation. This is kind of lengthy for some computers to handle when a shorter *still more accurate than sngwiz* evaluation would be more than enough for many situations. Would your AICM lose a significant degree of accuracy if the computing time were cut into half ?
I am wondering if you use some sort of Monte Carlo approach to solve each AICM situation over "millions of hands" ? If you do, it would be nice to allow the user some control over the calculation time (think pokerstove and/or a keyboard input). Currently, your AICM only gives a complete *perfect* evaluation. This is kind of lengthy for some computers to handle when a shorter *still more accurate than sngwiz* evaluation would be more than enough for many situations. Would your AICM lose a significant degree of accuracy if the computing time were cut into half ?
I do have some ways to reduce computation at the expense of some accuracy (some of these appeared in an earlier closed beta), but they probably wont make an appearance anytime soon. There's two reasons for this:
First is that there is still room to improve performance without affecting the results... especially by making the eventual move to GPU computing. I have several performance improvements planned for imminent versions.
Second is that I'm reluctant to add anything that complicates the user experience, at least for now. From all the feedback I've received so far, I've learned that one of the biggest hurdles I have to overcome is how to educate the community about what SnG Solver is all about.
I've gotten tons of emails that go something like this: "I know sngwiz and holdemresources are 100% correct so if SnG Solver says something different, it must be wrong!".. or.. "I dont believe it"... or ... "Holdemresources already calculates Nash, so how this can be better?"
In trying to overcome these kinds of preconceptions, I feel that anything that might complicate the results, or complicate the process of interpretation is just going to make my uphill climb that much steeper.
The poker community has been riding around on horseback for a long time now, and I've just showed up with the first automobile... change wont happen overnight but history will be the ultimate judge and I'm feeling confidant.
Quote:
Also, any thoughts on a release date as I can't find that info anywhere ! Thanks !!! This all looks interesting
Also, any thoughts on a release date as I can't find that info anywhere ! Thanks !!! This all looks interesting
I would like to release version 1.0 before the new year. The retail release will have a free-trial period and I will have a reduced "introductory price" for at least a month.Thanks for the interest! I always appreciate feedback of any kind.
12-17-2011
, 08:45 PM
12-18-2011
, 03:18 AM
12-18-2011
, 12:23 PM
Hey guys... there's a new version available (0.9.17) in the usual place: http://www.sngsolver.com/beta.html
Most of the whats new are behind-the-scenes fixes... but there is one pretty significant addition to the EV graph. Introducing: EV-RMSD! Or, more specifically, EV RMSD(root-mean-square deviation) with respect to uncertainty of opponent ranges.
It looks like this:
The darker shaded area surrounding the bright green Hero's all-in EV line represents what our EV might be if the ranges that we've assigned to our opponents aren't *exactly* correct.
Putting an opponent on a range can be tricky business... we might figure that an opponent will call our push with, say, 12% of hands. But can we really know for certain that his actual range isnt 10%? or 15%? And how much does it even matter if we're wrong? This is what the EV-RMSD band visualizes.
I tend to think of it as the "snap band". As in... if the green band is completely above the red fold equity line, then snap-shove! Or if its completely below the red fold equity line , then snap-fold! If the red line is somewhere within the EV-RMSD band then our situation is sensitive to our opponents behavior and its probably a spot where we should take the extra time to consider other factors that aren't so easily contained within the math... like our read on our opponents and that kind of thing.
Right now, the opponents likely distribution error (aka, "sigma") is set at +/- 20% of the nominal range. So, the sigma for a range of 10% is 8-12% (10 +/- 2%), or for range of 60% is 48-72% (60 +/-12%). I thought this was a reasonable starting point for a model of range uncertainty, but I'm open to suggestions for a better method. Of course, ultimately, this kind of thing could be made customizable by the user (I know that was going to be the first question asked).
Anyways, there you go. Enjoy.
Most of the whats new are behind-the-scenes fixes... but there is one pretty significant addition to the EV graph. Introducing: EV-RMSD! Or, more specifically, EV RMSD(root-mean-square deviation) with respect to uncertainty of opponent ranges.
It looks like this:
The darker shaded area surrounding the bright green Hero's all-in EV line represents what our EV might be if the ranges that we've assigned to our opponents aren't *exactly* correct.
Putting an opponent on a range can be tricky business... we might figure that an opponent will call our push with, say, 12% of hands. But can we really know for certain that his actual range isnt 10%? or 15%? And how much does it even matter if we're wrong? This is what the EV-RMSD band visualizes.
I tend to think of it as the "snap band". As in... if the green band is completely above the red fold equity line, then snap-shove! Or if its completely below the red fold equity line , then snap-fold!
If the red line is somewhere within the EV-RMSD band then our situation is sensitive to our opponents behavior and its probably a spot where we should take the extra time to consider other factors that aren't so easily contained within the math... like our read on our opponents and that kind of thing.Right now, the opponents likely distribution error (aka, "sigma") is set at +/- 20% of the nominal range. So, the sigma for a range of 10% is 8-12% (10 +/- 2%), or for range of 60% is 48-72% (60 +/-12%). I thought this was a reasonable starting point for a model of range uncertainty, but I'm open to suggestions for a better method. Of course, ultimately, this kind of thing could be made customizable by the user (I know that was going to be the first question asked
).Anyways, there you go. Enjoy.
12-18-2011
, 11:40 PM
12-22-2011
, 10:18 PM
12-23-2011
, 12:42 PM
Okay guys, version 0.9.18 is up!: http://www.sngsolver.com/beta.html
A few goodies in this update...
- The "Details" page is back with lots of numbers. Its still a work in progress and not everything is fully interactive yet, but you can see where its going.
- The bug ginandbread mentioned above is fixed (this was actually an old bug that I managed to re-create).
- The vertical scale on the EV graph is now adjustable... either from the menu or by hotkey (ctrl +/-)
- I believe I have fixed a problem that some people were having with activating their beta key. If you had trouble and got a "Code 6", you should give this version a try.
And my personal favorite...
- Added a "!" to the Push/Fold recommendation if the RoI exceeds the EV-RMSD threshold. Now you can know at-a-glance if SnG Solver thinks its okay for you to write "snapshove!" when you post in the STT Strategy forum.
Go nuts.
A few goodies in this update...
- The "Details" page is back with lots of numbers
. Its still a work in progress and not everything is fully interactive yet, but you can see where its going.- The bug ginandbread mentioned above is fixed (this was actually an old bug that I managed to re-create).
- The vertical scale on the EV graph is now adjustable... either from the menu or by hotkey (ctrl +/-)
- I believe I have fixed a problem that some people were having with activating their beta key. If you had trouble and got a "Code 6", you should give this version a try.
And my personal favorite...
- Added a "!" to the Push/Fold recommendation if the RoI exceeds the EV-RMSD threshold. Now you can know at-a-glance if SnG Solver thinks its okay for you to write "snapshove!" when you post in the STT Strategy forum.
Go nuts.
12-29-2011
, 04:38 AM
Hey guys.
Well... its time. Time to bring the beta test to a close. Of course there's still lots to be done, but its would be all too easy to just be in "beta" forever... sometimes you've got to just rip the band-aid off.
I had wanted to get hand-history import in before the "retail" release, but its not going to happen. But, rest assured that I will not end my "introductory price" until HH import is in.
I want to give a big thanks to all of you that took the time to download and try out SnG Solver and gave feedback... the response was fantastic and you've been a tremendous help.
If you didnt get a chance to try out the beta, dont worry... SnG Solver will have a fully functional 30-day free trial period. (available soon!)
On a slightly related note...
I'm pretty sure that I'm going to rename "Advanced ICM". As I alluded to in an earlier post, its become clear to me that the better I am able to educate people about SnG Solver's simulation-based equity algorithm, the more likely I am to be successful. To this end, I feel that I should probably put as much distance between me and the term "ICM" as possible.
At first, I thought it would be a good idea to latch on to the "ICM concept", only improved (ergo, "Advanced ICM")... but now I think that just causes confusion. Plus, there are already other programs that refer to themselves as "an advanced ICM calculator" and that kind of thing. So, let them have it, i say. I can do better.
So I say all this for two reasons... first, in case somebody wants to talk me out of it. and two, I'm open to ideas for the new name. In fact, if someone were to suggest a name and I use it, I'll give them a free SnG Solver license.
I'm thinking along the lines of "Simulated Chip Model".. or "Deep Simulation Equity Alogrithm"... or "Simulated Future Chip Model"... or something like that. You get the idea.
Well... its time. Time to bring the beta test to a close. Of course there's still lots to be done, but its would be all too easy to just be in "beta" forever... sometimes you've got to just rip the band-aid off.
I had wanted to get hand-history import in before the "retail" release, but its not going to happen. But, rest assured that I will not end my "introductory price" until HH import is in.
I want to give a big thanks to all of you that took the time to download and try out SnG Solver and gave feedback... the response was fantastic and you've been a tremendous help.
If you didnt get a chance to try out the beta, dont worry... SnG Solver will have a fully functional 30-day free trial period. (available soon!)
On a slightly related note...
I'm pretty sure that I'm going to rename "Advanced ICM". As I alluded to in an earlier post, its become clear to me that the better I am able to educate people about SnG Solver's simulation-based equity algorithm, the more likely I am to be successful. To this end, I feel that I should probably put as much distance between me and the term "ICM" as possible.
At first, I thought it would be a good idea to latch on to the "ICM concept", only improved (ergo, "Advanced ICM")... but now I think that just causes confusion. Plus, there are already other programs that refer to themselves as "an advanced ICM calculator" and that kind of thing. So, let them have it, i say. I can do better.
So I say all this for two reasons... first, in case somebody wants to talk me out of it.
and two, I'm open to ideas for the new name. In fact, if someone were to suggest a name and I use it, I'll give them a free SnG Solver license. I'm thinking along the lines of "Simulated Chip Model".. or "Deep Simulation Equity Alogrithm"... or "Simulated Future Chip Model"... or something like that. You get the idea.
12-31-2011
, 04:50 AM