Poker discovery software

Last time I made poker software I posted here and got incredibly valuable feedback. So even if I get curb stomped like this young man it's still worth it to try and answer the question: can this be usefull?

Video for TLDR http://assaultware.com/videos/rm-2/rm-2.html


I tried for several years to become an awesome poker player and while I did make some money, I never became even close to as good as some people who put in the same or even less effort. Why was this so? I constantly struggled with situations where I wasn't sure what to do.

Some simple examples:
-What are the best flops to cbet against fish (for value/for fold equity)?
-Is this a good scare card to bluff on?
-This guy just 2x pot on the river and I have 2 pair. What range could he have and is it a profitable call?

Posting hands on forums gets you mixed results and rarely a better explanation than "call", "easy shove", "you shouldn't be playing this high if you don't know this".

Watching videos didn't help me that much either. They would say things like "this is a great spot to cbet" but what's so great about it? There are so many variables and only few instructors give any explanation at all.

It occurred to me that all the data I needed had to be in my database somewhere. The summation of all the hands I've played should give me at least some idea of people's ranges and tendencies. Alas, none of the major trackers present aggregate information in any useful way - in fact even the new generation of trackers don't.

So I set out to make my own software that will help you answer these questions based off the hands you've actually played. No biased memory - pure data. The name of my new software is RängeMeister.

Here are some screen shots:


Analyze original or modified versions of any hands you've played


Get a listing of all the similar hands in your database


How could the hand have played out and how does my range/equity stack up in each spot?


Do quick what if analysis on ranges and reaction probabilities based on things like board texture, player types, and even specific stats


The goal of this software is to make it so you can get a clear view of spots that you face frequently or even rarely. Best case scenario is someone who is dedicated levels the playing field with those who have super human unbiased memory.




The software is currently available as free beta and can be downloaded from here: http://www.assaultware.com/downloads.aspx

RIght now it works only with HM/HM2. PT4 might be added but not PT3

downloading, this seems very good for videos, maybe too elaborate for personal analysis

this idea occurred to me a long time ago.. glad to see someone with the expertise and drive developing the idea and working on it.

wish I could mess around with it but I don't have my old databases or even access to HEM on this computer fk
this gives me impetus to get my desktop online

You say no biased memory, just pure data. If you're using your own hand histories it's going to be biased. When you start trying to figure out other players it's biased by the way they play against you and you run into heteroscedasticity in analysis when you begin trying to predict ranges. I think you're heading in the right direction, but just off the road a bit. One thing I would try to step away from completely is placing players into different categories and trying to examine how groups of players would approach a specific scenario. You're going to find some correlations for sure, but how much of that is just proper adjustment based on the differentiation in ranges they had going into that scenario in the first place? Don't get me wrong. I believe you still need to compare all players and find relations, but the math just gets more complex. Think about it like this: If you were looking for a particular tendency of a particular player you could take all the verifiable stats you have on that player and compare to every other player in your database with verifiable stats (not similar, just statistically significant). If you have 10,000 such players you have a system of 10,000 polynomial equations which you can then run regressions on and get an extremely accurate value for the tendency you're looking for. You don't need to group similar players together, you can learn a lot about how a particular TAG player may play something by using all the combined data including all of the different fish.

damn this looks better then CardRunners EV calculator

On February 02 2012 13:19 jchysk wrote: You say no biased memory, just pure data. If you're using your own hand histories it's going to be biased. When you start trying to figure out other players it's biased by the way they play against you and you run into heteroscedasticity in analysis when you begin trying to predict ranges.

maybe I'm misunderstanding you here but this is not just how people play you. It is any hand in your database - even if you weren't in the hand and even if you folded UTG pre flop then the blinds got in a war that hand would be used for searching blind v blind situations potentially.

I think you're heading in the right direction, but just off the road a bit. One thing I would try to step away from completely is placing players into different categories and trying to examine how groups of players would approach a specific scenario. You're going to find some correlations for sure, but how much of that is just proper adjustment based on the differentiation in ranges they had going into that scenario in the first place? Don't get me wrong. I believe you still need to compare all players and find relations, but the math just gets more complex. Think about it like this: If you were looking for a particular tendency of a particular player you could take all the verifiable stats you have on that player and compare to every other player in your database with verifiable stats (not similar, just statistically significant). If you have 10,000 such players you have a system of 10,000 polynomial equations which you can then run regressions on and get an extremely accurate value for the tendency you're looking for. You don't need to group similar players together, you can learn a lot about how a particular TAG player may play something by using all the combined data including all of the different fish.

You can create your own groups and even search just based on a couple of stats. To some extent the software would only work as well as the user makes it work but the flexibility you mention is there

Looks really nice. The value of datamined hands goes up after this, I'd say.
Question you arguably didn't expect: What did YOU learn from playing with it? (care to elaborate a little?)

You should make a vid of "how to use"

On February 02 2012 15:45 Baalim wrote: You should make a vid of "how to use"

Indeed. like the flopzilla one: http://www.thepokerbank.com/videos/splitsuit/using-flopzilla/
Also I tried this a while ago, and it was a major hassel to get the right version of .net framework. I think I downloaded one version, then another, then another. Ideally would be good to have the correct .net in with the download even if it make rangemister a 100 meg download. or failing that, a super direct link to the correct .net package, so that the user only has to do one click to insall it. not a link to a webpage that microsoft keeps chainging.
Its a great idea, and the smiling german icon is classy

On February 02 2012 15:26 anheway wrote: Looks really nice. The value of datamined hands goes up after this, I'd say. Question you arguably didn't expect: What did YOU learn from playing with it? (care to elaborate a little?)

This is an unexpected question and a very good one. I haven't played poker basically at all for awhile and I think to get any value from this I would have to start playing again at least a few thousand hands per month. In fact I will be doing this as to really try and evaluate both my software applications from the perspective of a player. I am posting here first because I am hoping that people who play much better than I do (baal for example) will give feedback like "this sucks for the following reasons..." and I can adjust it accordingly before diving in myself.

On February 02 2012 15:45 Baalim wrote: You should make a vid of "how to use"

http://assaultware.com/videos/rm-2/rm-2.html

Lots of features and fixes since that video was made - most notably the graphing options when you right click a node.

This brings up an interesting question. There's a link on the "start" page for this video. One of two things is happening here:

1. It is not in a place where it is easy enough to see
2. The video wasn't helpful

I am thinking it's more likely #1 because I have those same icons all over the place in my other software and I still get tons of questions that would be answered by the videos.

Indeed. like the flopzilla one: http://www.thepokerbank.com/videos/splitsuit/using-flopzilla/ Also I tried this a while ago, and it was a major hassel to get the right version of .net framework. I think I downloaded one version, then another, then another. Ideally would be good to have the correct .net in with the download even if it make rangemister a 100 meg download. or failing that, a super direct link to the correct .net package, so that the user only has to do one click to insall it. not a link to a webpage that microsoft keeps chainging. Its a great idea, and the smiling german icon is classy

This is great feedback and I very much appreciate it. On my old site I had a link to the right framework version but I completely ignored putting it on the new site. Adding it to the installer seems like a great idea

I designed the logo myself though someone else drew it

On February 02 2012 14:38 Roald wrote: Show nested quote + On February 02 2012 13:19 jchysk wrote: You say no biased memory, just pure data. If you're using your own hand histories it's going to be biased. When you start trying to figure out other players it's biased by the way they play against you and you run into heteroscedasticity in analysis when you begin trying to predict ranges. maybe I'm misunderstanding you here but this is not just how people play you. It is any hand in your database - even if you weren't in the hand and even if you folded UTG pre flop then the blinds got in a war that hand would be used for searching blind v blind situations potentially.

You're right, but I'm thinking more about ranges. Any situation where there's hole cards being taken into consideration which are usually only seen during showdowns you're getting a skewed dataset to work with. The numbers might be pure, someone is 3betting X percentage from Y position with Z chips, but you don't know what the distribution of hands is underneath.

On February 02 2012 14:38 Roald wrote: Show nested quote + I think you're heading in the right direction, but just off the road a bit. One thing I would try to step away from completely is placing players into different categories and trying to examine how groups of players would approach a specific scenario. You're going to find some correlations for sure, but how much of that is just proper adjustment based on the differentiation in ranges they had going into that scenario in the first place? Don't get me wrong. I believe you still need to compare all players and find relations, but the math just gets more complex. Think about it like this: If you were looking for a particular tendency of a particular player you could take all the verifiable stats you have on that player and compare to every other player in your database with verifiable stats (not similar, just statistically significant). If you have 10,000 such players you have a system of 10,000 polynomial equations which you can then run regressions on and get an extremely accurate value for the tendency you're looking for. You don't need to group similar players together, you can learn a lot about how a particular TAG player may play something by using all the combined data including all of the different fish. You can create your own groups and even search just based on a couple of stats. To some extent the software would only work as well as the user makes it work but the flexibility you mention is there

Your software is really neat. It makes things far easier that would have only been possible by writing some pretty intense queries directly to the HM or PT database. I think I may have been approaching a different problem than what your software is built to handle. It seems to be figuring out the most equitable choice based on the information you have is the easy math. The difficult part is getting accurate and valuable information with which to deduct those choices. As to the example I gave previously, you could have your database of players with stats like VPIP, PFR, 3Bet, etc. but a lot of stats that are near useless for most players like Cbet River or Check Raise turn because there just aren't enough situations where those instances occur for an individual player. What you can do though is take the collective of all your players and figure out pretty strong values for those stats given all their other stats like the VPIP and PFR. The example I gave above or use of Bayesian modelling are two possible avenues I could see taking.

Good thread so far and looks interesting. Will download and play around with it when I get a second.

edit: doesn't matter

Cool. Does it work for PLO too?

On February 02 2012 17:42 jchysk wrote: Show nested quote + On February 02 2012 14:38 Roald wrote:   On February 02 2012 13:19 jchysk wrote: You say no biased memory, just pure data. If you're using your own hand histories it's going to be biased. When you start trying to figure out other players it's biased by the way they play against you and you run into heteroscedasticity in analysis when you begin trying to predict ranges. maybe I'm misunderstanding you here but this is not just how people play you. It is any hand in your database - even if you weren't in the hand and even if you folded UTG pre flop then the blinds got in a war that hand would be used for searching blind v blind situations potentially. You're right, but I'm thinking more about ranges. Any situation where there's hole cards being taken into consideration which are usually only seen during showdowns you're getting a skewed dataset to work with. The numbers might be pure, someone is 3betting X percentage from Y position with Z chips, but you don't know what the distribution of hands is underneath.

There is always the problem of the cards only being known if they were shown down and this definitely does matter. However, that is true even if you're just playing. I think what makes up for it is that you can run through the whole hand and can figure out fold equity for various lines which helps you make determinations even if the cards end up being unknown.

Show nested quote + On February 02 2012 14:38 Roald wrote:   I think you're heading in the right direction, but just off the road a bit. One thing I would try to step away from completely is placing players into different categories and trying to examine how groups of players would approach a specific scenario. You're going to find some correlations for sure, but how much of that is just proper adjustment based on the differentiation in ranges they had going into that scenario in the first place? Don't get me wrong. I believe you still need to compare all players and find relations, but the math just gets more complex. Think about it like this: If you were looking for a particular tendency of a particular player you could take all the verifiable stats you have on that player and compare to every other player in your database with verifiable stats (not similar, just statistically significant). If you have 10,000 such players you have a system of 10,000 polynomial equations which you can then run regressions on and get an extremely accurate value for the tendency you're looking for. You don't need to group similar players together, you can learn a lot about how a particular TAG player may play something by using all the combined data including all of the different fish. You can create your own groups and even search just based on a couple of stats. To some extent the software would only work as well as the user makes it work but the flexibility you mention is there Your software is really neat. It makes things far easier that would have only been possible by writing some pretty intense queries directly to the HM or PT database. I think I may have been approaching a different problem than what your software is built to handle. It seems to be figuring out the most equitable choice based on the information you have is the easy math. The difficult part is getting accurate and valuable information with which to deduct those choices. As to the example I gave previously, you could have your database of players with stats like VPIP, PFR, 3Bet, etc. but a lot of stats that are near useless for most players like Cbet River or Check Raise turn because there just aren't enough situations where those instances occur for an individual player. What you can do though is take the collective of all your players and figure out pretty strong values for those stats given all their other stats like the VPIP and PFR. The example I gave above or use of Bayesian modelling are two possible avenues I could see taking.

That starts to get heavily into artificial intelligence and I would like to eventually evolve the software more in that direction. I think potentially it could try different stat filters and see what causes the most deviations from the unfiltered data while still having a significant sample size and might give some interesting results.

On February 03 2012 15:17 Joe wrote: Cool. Does it work for PLO too?

Not right now but if people think the holdem version is usefull I could easily add it since this runs off the same engine as my other software which does support plo