ICM-ish idea: Calling ranges

[bogey1]

So, I've got hundreds of 6-man SNGs sitting in my hand history directory. One of my difficulties with ICM is putting people on a range with any accuracy. I'm not really sure my gut feel for their range during post game analysis is really any better than my gut feel for whether to shove/fold during the game.

So, I had a thought. I'd write a script to go through and look for any preflop all-in followed by a call, then extract the hand that called from the showdown output.

This should give me a few thousand examples of what people called with. Obviously there will be some tainting of the data from people calling micro-stacks or multiple callers (e.g. raise, reraise, call 3-way pots). I can track how many times each hand called to help with that though, such as seeing AK called 12 times but 95o only once. Based on frequency and hand strength I can probably figure out how to throw out the oddball calls. For example, if AK called 10 times, and AJ 8 times, but AQ 1 time, based on hand strength AQ goes in the range, but that 95o that called once would be thrown out as both rare and low hand strength.

Has anyone done something like this before? Do folks think the calling ranges I come up with would be in the least bit useful?

I got as far as extracting the showdown hands from preflop all-in confrontations in about 20 minutes. Taking that data and massaging it into the call range may take a little bit though.

[AMT]

not familiar with anyone trying to do this, not even sure i fully understand this, the problem is i just dont see, above consolidation and tracking of the actual hands for analysis purposes, that this would help you develop an understanding of complete hand ranges and situations that would help you decide how to read opponents ranges...

also, whose to say these ppl are calling with reasonable frequencies and hands that fare well against supposed ranges? they may be calling way too tight/too loose and you wouldnt know how to adjust for the hand ranges that youd be facing...i try not to put my faith in random players at the game i play bc i think they all suck. just a couple of problems i see with it....

[PokahPokah]

sorry hijack op`s thread but to continue with this idea what do you all use to put people on ranges, stack sizes etc ? im sucking at this part too at the mo

[Slim Pickens]

PokerTracker filters... and you'd probably need 10k SNGs to anything close to a useful sample.

[pifhluk]

[ QUOTE ]
sorry hijack op`s thread but to continue with this idea what do you all use to put people on ranges, stack sizes etc ? im sucking at this part too at the mo

[/ QUOTE ]

Notes
Stack Sizes
Position
Level of SNG
Multitabler

It is really important to take notes especially on the people that you play with alot, the regulars.

[bogey1]

The point, I suppose, is you need to put villians on a calling range to decide what's a profitable push range.

How are we deciding this range now on essentially unknown villians or when shoving into 2 or 3 villians? We're guessing based on...gut feel? I can put in numbers to SNGPT and run ICM tests, but I really don't have feedback that ever tells me if I'm characterizing the calling range correctly. Given we're often talking about taking small edges (+.3% type of things) it's very easy to make minor errors in the assumed hand ranges that move things from one side to the other of that profitability line.

What I'm thinking is that if I get a list of say 1000 hands that called an all in, then I actually have some concrete data on calling ranges.

I still have sampling issues for individual hands, but I can take the data and plot it by hand strength to smooth things out. Here's an attempt at an ascii graphic as I have no where to host an image.
<font class="small">Code:</font><hr /><pre>
#|
|
o|
f|
|
o|
c| x
c| xxx
u| xxxxx
r| xxxxxxx
e| xxxxxxxxxx
n| xxxxxxxxxxxxxx
c| xxxxxxxxxxxxxxxxxxx
e| xxxxxxxxxxxxxxxxxxxxxxxxx
s|===========================
0 10 20 30 40 50 60 70 80 90
S/C value
</pre><hr />

The point here is we can see the calling range starts around the S/C value of 30. From there I can get a hand range (basically, all hands of SC 30+). From that, I can plug that range into SNGPT and reverse it to get the proper shoving range against that calling range.

(These are completely fabricated numbers as an example).

Obviously, villians vary, but like a starting hand chart, it gives you base from which to shift up or down on your shoving range. Shift tighter the more opponents yet to act, shift looser if the BB is a loose caller, etc.

[Guthrie]

I've been toying with this for some time. The problem is that there a great many variables to consider. What a player calls with when you both have 10BBs and there are 7 players remaining is quite different than what he'll call with if he has 30 and you have 10 in the same situation, and different than what he'll call with on the bubble, again, with various stack sizes.

If you ignore the variables, then you end up with very dangerous results. Someone calling with 84o when he had 40BB and you had 2, will fatally taint your data.

[bogey1]

[ QUOTE ]
If you ignore the variables, then you end up with very dangerous results. Someone calling with 84o when he had 40BB and you had 2, will fatally taint your data.

[/ QUOTE ]

Yea, the data would,in my case, be pretty specific to 1-table 6-man SNGs. Most of the data is going to come from the point where there's 2-4 people. I could fairly easily restrict it to just 3-4 people to avoid HU and very early all-ins from throwing off the results.

At 3-4 players, position becomes fairly irrelevant for calling a shove. A shove from UTG/CO isn't much tighter than a shove from the BU or SB.

But, yea, data tainting is a problem. I'm hoping the graphing by occurence in hand strength ranges will help. That 84o call...well, it's going to happen so rarely it'll be clear from the curve in the graph that it's an outlier and not something I'd consider as part of the "normal" calling range.

[plexiq]

The problem really is: You would need huge amounts of data (or alternatively some intelligent way to generalize from your restricted data).

If you do it straightforward, you basically have 2 ways:
1) Simply average over all kinds of different situations. (Ignoring stack sizes, # of players left, blinds, etc) That will lead to awful accuracy, obviously. (See example in the last post.)

2) Use only data that closely matches your current situations. As an example, you have a situation of Blinds 200/400 and stacks 1000, 5000,2000,1000,3500, and you want to know the Calling range of BB vs a Push from the CO. Even in huge databases, you will likely have problems to find many hands where Stacks match approximately AND Blinds match AND the CO pushes AND the BB calls (so we see a showdown). If we are lucky, we will find a few hands that match the criteria - but nowhere enough for a meaningful sample.

I think using some approximated Nash Equilibrium range as baseline, and adjusting it for the buyin-level/specific villain is much easier and probably gives much better results.

[BradleyT]

This will be extremely hard to do.

Let's say you put me on 22+,A2+,K7+,Q8s+,Q9o+,JT+ but I know my range is really 66+,AQ+,KQ+. Your hand is a call against the range you think I'm on however it's a huge fold against my actual range. In this case I have 66 and push and you think you made a great profitable call. Next time I have AQ and you have AK and think you made a brilliant call even though again it was incorrect vs. my range. You'd have to repeat this exact hand hundreds of times to ensure I was dealt every hand at least once before you could pinpoint my exact range.

Edit - Actually you'd have to repeat it probably thousands of times since many times you could not call to see my cards.