Skip to forum
Midlife Crisis? Wha...
 
Notifications
Clear all

Midlife Crisis? What Midlife Crisis?

8,661 Posts
251 Users
37 Reactions
787.6 K Views
jbpatzer
Joined: 23.11.2009

Originally posted by jules97

Originally posted by jbpatzer
I tried these ranges with a single bet size of 0.75 pot, and it takes four minutes to do what I regard as one complete CFRM sweep, that is going through all possible combinations of flop buckets, turn card buckets and river card buckets, of which there are many. It's not entirely clear to me how many of these sweeps I need to get a solution that's close enough to optimal to be useful, but I think its at least 1000, probably 5000. That means this set of ranges is probably just out of reach at the moment. Once I have the code multithreaded and run on a multicore machine, I think it will be doable, which seems reasonable to me given what I've read on the state of the art.

In the meantime, would you like to suggest a less challenging problem that you'd be interested in solving? How about BTN raises, BB 3bets, BTN calls? If you give me some ranges for that, I can give it a go.

Ok no problem,
I think 3b pots with BTN are equally important anyway.

Maybe it's best to start with a linear 3b range from SB rather than a BB range.

Here's a SB 3b range 21%
AA-44,AKo-ATo,KQo-KTo,AKs-A2s,KQs-K8s,QJs-Q9s,JTs-J8s,T9s-T8s,98s-97s,87s-86s,76s-75s,65s

Here's a BTN CC range: 20%
TT-44,AJs-A2s,KQs-K7s,QJs-Q8s,JTs-J8s,T9s-T8s,98s-97s,87s-86s,76s-75s,65s,AQo-A9o,KQo-KTo,QJo

Lets make a 2bb open and 8bb 3b. So that would the pot at 17bb. So I think a 50% pot & shove option would probably be best. That would leave a 60% shove on the river when it goes bet,bet.

OK. I'm just finishing off Janda Hand 1 with 0.25, 0.5, 0.75, 1, 1.25 and 1.5*pot bet options and I'll get onto it.


Reply
Quote
jules97
Joined: 10.06.2012

There was also a hand recently on the hand evaluation forums.
NL100 KQs Speed

A 3b pot between CO and MP calls the 3b. The flop comes K:hJ:s8:h and a question was is x/r JJ on the flop as MP better than calling? There was a bit of discussion, and I said something like 'herp derp, not raising JJ on the flop is losing money, raising is good, derp' but on second thoughts, I really have no idea.

So maybe if you ran some ranges and posted it all smug and obnoxious like, it might drum up some more interested folks. Ranges are fairly tight, so maybe it won't take too long.

Here's some generic ranges, RFI to 3b, 3b to 9bb
CO 3b range:
AA-QQ,AJo,KQo,AKs,A7s,A3s-A2s,KTs,QTs,J9s,T8s,87s, 50% AKo
and a MP cc range:
JJ-99,AQs-AJs,KQs,QJs,JTs,AQo


Reply
Quote
jbpatzer
Joined: 23.11.2009

Originally posted by jules97
There was also a hand recently on the hand evaluation forums.
NL100 KQs Speed

A 3b pot between CO and MP calls the 3b. The flop comes K:hJ:s8:h and a question was is x/r JJ on the flop as MP better than calling? There was a bit of discussion, and I said something like 'herp derp, not raising JJ on the flop is losing money, raising is good, derp' but on second thoughts, I really have no idea.

So maybe if you ran some ranges and posted it all smug and obnoxious like, it might drum up some more interested folks. Ranges are fairly tight, so maybe it won't take too long.

Here's some generic ranges, RFI to 3b, 3b to 9bb
CO 3b range:
AA-QQ,AJo,KQo,AKs,A7s,A3s-A2s,KTs,QTs,J9s,T8s,87s, 50% AKo
and a MP cc range:
JJ-99,AQs-AJs,KQs,QJs,JTs,AQo

50% AKo. Bah! You really want me to add the ability to do that?? Pff. Maybe.

EDIT: Lol. I'm too much of a fish to access that hand. I'll have to play some Zoom to get gold status. Not sure whether that counts as motivation or irritation... _ugly:


Reply
Quote
jbpatzer
Joined: 23.11.2009

Originally posted by jbpatzer

Originally posted by jules97

Originally posted by jbpatzer
I tried these ranges with a single bet size of 0.75 pot, and it takes four minutes to do what I regard as one complete CFRM sweep, that is going through all possible combinations of flop buckets, turn card buckets and river card buckets, of which there are many. It's not entirely clear to me how many of these sweeps I need to get a solution that's close enough to optimal to be useful, but I think its at least 1000, probably 5000. That means this set of ranges is probably just out of reach at the moment. Once I have the code multithreaded and run on a multicore machine, I think it will be doable, which seems reasonable to me given what I've read on the state of the art.

In the meantime, would you like to suggest a less challenging problem that you'd be interested in solving? How about BTN raises, BB 3bets, BTN calls? If you give me some ranges for that, I can give it a go.

Ok no problem,
I think 3b pots with BTN are equally important anyway.

Maybe it's best to start with a linear 3b range from SB rather than a BB range.

Here's a SB 3b range 21%
AA-44,AKo-ATo,KQo-KTo,AKs-A2s,KQs-K8s,QJs-Q9s,JTs-J8s,T9s-T8s,98s-97s,87s-86s,76s-75s,65s

Here's a BTN CC range: 20%
TT-44,AJs-A2s,KQs-K7s,QJs-Q8s,JTs-J8s,T9s-T8s,98s-97s,87s-86s,76s-75s,65s,AQo-A9o,KQo-KTo,QJo

Lets make a 2bb open and 8bb 3b. So that would the pot at 17bb. So I think a 50% pot & shove option would probably be best. That would leave a 60% shove on the river when it goes bet,bet.

OK. I'm just finishing off Janda Hand 1 with 0.25, 0.5, 0.75, 1, 1.25 and 1.5*pot bet options and I'll get onto it.

I just set this going. It's going to take 5 days. :f_mad: :f_zZz: :rage: :facepalm:


Reply
Quote
jules97
Joined: 10.06.2012

Oh snap :(

Surely you have the blog hits to be promoted permanently to blackmember.


Reply
Quote
jbpatzer
Joined: 23.11.2009

Originally posted by jules97
Oh snap :(

Surely you have the blog hits to be promoted permanently to blackmember.

I can't disagree with that.

If I can get it multithreaded, we'd be in business. I'm trying to learn exactly enough about how to do it to be able to get it going for precisely my code. This approach is great when it works, but when it doesn't, which it currently doesn't, I'm left scratching my head and going back to read more stuff about it.

If anybody would like to explain to me exactly how to code a parallel for loop in C++ for the bit of my code in the spoiler, which doesn't currently do anything other than crash MATLAB, I'd be eternally grateful. Thanks to my 'need to know' approach to this, I have no fkn clue what the majority of

Concurrency::parallel_for(1, *(lnc+j-1), [&] (int l)

means. :f_mad:

Spoiler

Concurrency::parallel_for(1, *(lnc+j-1),
[&] (int l)
{
if (*(cN+j+(l-1)*Ni-1))
{
if (*(showdown+ *(cmN+j+(l-1)*Ni-1) -1))
v[l-1] = *(winner+f11+*Nfb*(f21-1+*(Nfb+1)*(t-1+Ntcb*(r-1)))-1) *
*(weight+f11+*Nfb*(f21-1+*(Nfb+1)*(t-1+Ntcb*(r-1)))-1) *
*(sdvalue+*(cmN+j+(l-1)*Ni-1)-1);
else
v[l-1] = *(weight+f11+*Nfb*(f21-1+*(Nfb+1)*(t-1+Ntcb*(r-1)))-1)*
*(sdvalue+*(cmN+j+(l-1)*Ni-1)-1);
}
else
{
CFR(*(children+j+(l-1)*Ni-1),active, f11, f21, t, r,strat,regret,stratsum,
floptoturn1, floptoriver1, floptoturn2, floptoriver2, sdvalue, player,
children, lnc, showdown, winner, weight, cN, cmN, Nc,Ntcb,maxNrcb,Ni,Nfb,index,false,value0);
v[l-1] = *value0;
}
});

Note that this bit of code comes within the function CFR, which calls itself recursively.


Reply
Quote
roopopper
Joined: 31.12.2010

Can't make Zimmers tonight but deffo up for next week :) Have you set up new skype zimmers group?


Reply
Quote
jbpatzer
Joined: 23.11.2009

Originally posted by roopopper
Can't make Zimmers tonight but deffo up for next week :) Have you set up new skype zimmers group?

Roo: 'Can we play?! Can we play?! Pleeease! O! Actually I can't play.'
Me: :facepalm:

Only Gary and Vorpal have registered at the new club, although we could play in the old club, except Gary's not a member. I wonder whether anyone else will join. Idk. Maybe next week.

Started creating a new Zimmers Skype group as requested. Realised the old one was still there. :facepalm:


Reply
Quote
jbpatzer
Joined: 23.11.2009
jules97
Joined: 10.06.2012

Well that's a bummer. It uses blockers for bluff raises, so it looks to be using at least some removal effects.

That's the 2nd piece of bad news that's come from your direction. It appears jokerstars are under pressure to leave aus/russ markets due to this UK licence business. Rumour is Nov 1 may bring some bad news. Fingers crossed nothing comes of it. I have'nt played much there recently, but it would still be nice to have it as an option. :( Pommy bastards...


Reply
Quote
jbpatzer
Joined: 23.11.2009

Originally posted by jules97
Well that's a bummer. It uses blockers for bluff raises, so it looks to be using at least some removal effects.

That's the 2nd piece of bad news that's come from your direction. It appears jokerstars are under pressure to leave aus/russ markets due to this UK licence business. Rumour is Nov 1 may bring some bad news. Fingers crossed nothing comes of it. I have'nt played much there recently, but it would still be nice to have it as an option. :( Pommy bastards...

We are bastards....for sho.

I'll have to think about it, but I feel that my bucketting algorithm still gives solutions that must be close to equilibrium. The problem comes, as Alex pointed out, in river situations like

Ad2s3d8h9c

Player 1: AcKc, KcKs, KdKh, QcQs
Player 2: AcJc, KcKs, KdKh, QdQh

My algorithm buckets all KK together, but player 2's KcKs blocks Player 1's AcKc, whilst Player 2's KdKh does not. :( In this situation, my code would know that AcKc is blocked half of the time, and take this into account, but it loses the fine-grained information, 'KcKs = no AcKc, KhKd = one combo of AcKc', and instead takes the average.

If you go through and take this into account, you get far more hand buckets and the algorithm gets an order of magnitude slower.

However, these are very narrow ranges. If both players are instead playing AKs, KK and QQ, the frequency of player 1 having AKs on the river halves if Player 2 has KK, but this is taken into account. In situations where the suits matter for hand strength, the bucketting takes it into account, and then it does, as you point out, make use of blocker effects.

I think at the moment I'm leaning towards 'there is some loss of information, but the solutions are still very useful', but I'll give it some more thought.


Reply
Quote
jbpatzer
Joined: 23.11.2009
roopopper
Joined: 31.12.2010

Originally posted by jbpatzer

Roo: 'Can we play?! Can we play?! Pleeease! O! Actually I can't play.'
Me: :facepalm:

lol :f_biggrin:


Reply
Quote
roopopper
Joined: 31.12.2010

Originally posted by roopopper

Originally posted by mbml

Originally posted by roopopper
Did I win a book? :coolface:

Probably not :coolface:

Damn I was going to use the money I saved not buying your book to buy "education of a modern poker player" :coolface:

Sorry JB I can now not afford your book :f_cry:


Reply
Quote
jbpatzer
Joined: 23.11.2009

Originally posted by roopopper

Originally posted by roopopper

Originally posted by mbml

Originally posted by roopopper
Did I win a book? :coolface:

Probably not :coolface:

Damn I was going to use the money I saved not buying your book to buy "education of a modern poker player" :coolface:

Sorry JB I can now not afford your book :f_cry:

Probably not advanced enough for Platinum status sharks like you anyway. Save your money for Hellmuth's book.


Reply
Quote
roopopper
Joined: 31.12.2010

Originally posted by jbpatzer

Originally posted by roopopper

Originally posted by roopopper

Originally posted by mbml

Originally posted by roopopper
Did I win a book? :coolface:

Probably not :coolface:

Damn I was going to use the money I saved not buying your book to buy "education of a modern poker player" :coolface:

Sorry JB I can now not afford your book :f_cry:

Probably not advanced enough for Platinum status sharks like you anyway. Save your money for Hellmuth's book.

I spoke to TT a few days ago and he said I need to read the book :f_biggrin:


Reply
Quote
jbpatzer
Joined: 23.11.2009
jules97
Joined: 10.06.2012

I trust the drink a lot of wine plan didn't make you too sick. My plan to not drink a lot of beer on Saturday night could have gone better.

A♥7♣6♦

Had a look at the SB flop cbet, x/r, x/c & BTN b when checked to ranges.

SB

Cbetting 34% (28% half pot, 6% pot), which the total is almost identical to what I worked out before hand. Go me. :f_cool: Didn't really expect the 6% pot. But I imagine the EV difference is small, having this option. Also the range construction appears to be very similar as the half pot, only with different ratio's because of betsize. So I'm going to skim over this for now.
x/r only 6% No huge surprise here on this dry flop.

SB cbets most bdfd's and x/f's most of the suited spades. Mostly x/f those bdfd's it doesn't cbet if they don't have K high showdown value.

SB x/c's all gutshots when not cbet (or x/r) which is interesting, even T♠8:s with no fd.

Cbets 44 and 55 about a third of the time and x/c's them most of the time when not cbet
86s, 56s split 50/50 between x/c and cbet, the weakest 5's and 6's get x/r

PP's from 88-KK, prefers to x/c, sometimes cbets lower pp's. With increasing frequency the lower they get, up to 36% with 88's,
Very rarely ~5% of the time, the weakest of these pp's get x/r

Aces below AT get mostly x/c, the first Aces to get cbet are spades.
Then about 50% AT-Ak get cbet, spades more likely to get cbet, bd/fds more likely to get x/c (total combos cbet of AT-AK equivalent to about all combos of AK and AQ)

6♠7:s gets potted 85% x/r 15%, 6♥5:h similar but a little less aggressively

AA, cbet about 35% x/c the rest, A:sAx always x/c. I take it this is because non A:s aces are the suited ones with bdfd's BTN is more likely to bluff with.

66, 77 cbet ~75%, x/r ~ 8%, x/c the rest
77 has the most EV of any hand at 36.6bb, AA ~35.5bb, 66 34bb

BTN

Bets only half pot when checked to, 30% of the time. Checks behind the remaining 70%.
Bets most bd/fd's, a little happier to x behind K high ones, more happy the better they are
Interestingly it likes betting QJo also, 43%, KTo 29%, never KJo, never KQo. I assume they have good blockers, and less showdown value (unlike KJ, KQ)
44-55 5% of the time
weakest 5's ~70%
weakest 7's ~60%
similarly to BTN starts betting Aces from A9+
Bets 90% 2p+

At some stage I'll go over turns/rivers.

All of this seems in line with what I expected. The biggest thing to me is still the mixed strategies. Which as I've said before aren't the most common. I'm going to have to go away and have a good think about coming up with some simple, solid rules to implement some more mixed strategies in my game.
And also try and work out some exploitative adjustments and their benefits to people who aren't using mixed strategies (I'd say like 90%+ regs playing midstakes). I think these exploitative adjustments will do pretty well because it's one thing to balance ones ranges (which is hard enough!), but it's another to keep them balanced while playing mixed strategies (and dealing with them) and adjusting might not be so easy for someone who hasn't put in too much work off the table.


Reply
Quote
jbpatzer
Joined: 23.11.2009

I'm glad the results are still looking sensible. Hopefully I'll have coded up this epsilon distance thing soon, and then we'll know how exploitable these solutions are. Basically one player plays the computed strategy, and tells the other player what it is. The other player then chooses the maximally exploitative strategy at each decision node, and we calculate the EV of each player compared to the computed equilibrium EV. I don't expect the solution to be very exploitable at all, since, as far as I can tell, the only information loss is some small errors in hand v hand blockers effects on frequencies, but you never know.


Reply
Quote
jules97
Joined: 10.06.2012

yeah I have a feeling it's not going to be a huge deal

Ben Sulsky is on the latest thinking poker podcast if u havent seen it. Talks a little bit about game theory and stuff, pokersnowie, neural nets and even mention CFRM. Nothing too detailed, but I found it interesting. Apparently that limit bot hyperborean that no one can beat is CRFM too.


Reply
Quote