Originally posted by jbpatzer
Crazy Argentinian is rock climbing in Yosemite Park. She's got worse since you last spoke to her.
lol I remember seeing some pics of rock climbing on her facebook! 
Notifications
Clear all
J:c6:h4:d
After seeing a cbet, JBR's BTN raises 5%
Majority of value low sets, 25% of 44 and 25% of 66. Only 6% of top sets. Which makes sense. For bluffs it's going with the set blockers mainly. We're seeing a pattern here. Set blockers for bluffs are good. On dryish boards, with not so wide ranges.
Then it has the smattering of gutshots, overcards/bdfd's etc.
On to the 8:h8:d6:c MP v BTN hand
MP has a decent equity and equity distribution over BTN.
By the way, I'm getting different equity values when I put Janda's preflop ranges into CREV. It's very possible I'm making a mistake entering ranges or something, but this is the second time now I've noticed this, so maybe there's more to it. If I continue to notice, Ill mention it again. Not terribly important.
I get 52.1% equity for MP with CREV, JBR has 50.5%
Janda has no raising range as BTN. Neither does JBR. JBR's BTN calling 66% and Janda 66%t too.
There are however differences in range construction. Janda is calling all his BDFD's. All stuff from AJs to J9s. Janda also is calling AQo.
JBR instead prefers to fold these hands the majority of the time, instead calling hands PP's. All 77's, ~70% 55's, 50% 44's.
I explain this as, Janda's ranges were prefered at the time (like his preflop ranges), and maybe still some do currently, due to people's broken checking and barreling ranges, where x/f too often happens.
Floating with a hand like AQ (or overcards bd/fd) is done because whilst it may be behind a villains value betting range, it has equity against it, can outdraw and importantly it has an exploitably profitable bluff later on against someone not defending their checking range well. I think a hand like 44 (and especially 77) make more sense to defend instead when MP has solid checking ranges because they have higher equity and there will be no chance to exploitably bluff later on.
As MP
Janda Cbets 34%
x/c's 37% , x/r's 13%, xf's 50%
JBR cbets 45%
x/c's 1%!!!, x/r's 30%!!!, x/f's 69%
I didn't expect that! I don't see players do this at all.
Let's just ignore that calling range and call it 0.
For value it's raising 99+ 100% of the time.
Raising OESd's it gets there with, gutshots, 3 to a straights, nut bd/fds, some other bd/fd's (K:hT:h stuff). All the set blockers it gets here with (76s stuff)
90% of 25% it sees this spot with of AK combos. So I guess that's not that many. But still... interesting. I suppose they make better raises than folds.
A quick scan of BTN's defending range after seeing this x/r as expected shows BTN defending some very marginal hands. 50% of the time 44. All AK. ATs with BD/fd. Some KQ with bd/fd. ANd very, very rarely (15% of 41%) J:d9:d!
I put 886 hand into pokersnowie and got total different result. It might be due to diferet PF ranges MP or is loose 44+, A2s+, K7s+, Q9s+, J9s+, T8s+, 98s, 87s, ATo+, KTo+, QTo+, JTo; and BU call is tight: TT-55, AQs-AJs, KJs-KTs, QTs+, J9s+, T9s, 76s, 65s, 54s, AQo. There some mixed strategy with calls but not so much, could you put those ranges and we'll see how much snowie is of. Now it cbets only 13% and ch/f 70% which seems way too tight.
Originally posted by jules97
On to the 8:h8:d6:c MP v BTN hand
MP has a decent equity and equity distribution over BTN.
By the way, I'm getting different equity values when I put Janda's preflop ranges into CREV. It's very possible I'm making a mistake entering ranges or something, but this is the second time now I've noticed this, so maybe there's more to it. If I continue to notice, Ill mention it again. Not terribly important.
I get 52.1% equity for MP with CREV, JBR has 50.5%
I get exactly 50.5% from Equilab, so I think you must be entering the ranges wrong.
I've now modified the code to be able to do turn and river situations as well. I just need to do a bit more work tonight, and I'll test it on some obvious cases, and also run it on GTO Brainteaser #8 in the GTORangeBuilder blog so that I can post the solution there like the smug, obnoxious bastard that I am.
Originally posted by dooleslovs
I put 886 hand into pokersnowie and got total different result. It might be due to diferet PF ranges MP or is loose 44+, A2s+, K7s+, Q9s+, J9s+, T8s+, 98s, 87s, ATo+, KTo+, QTo+, JTo; and BU call is tight: TT-55, AQs-AJs, KJs-KTs, QTs+, J9s+, T9s, 76s, 65s, 54s, AQo. There some mixed strategy with calls but not so much, could you put those ranges and we'll see how much snowie is of. Now it cbets only 13% and ch/f 70% which seems way too tight.
Interesting. So just to be clear, Snowie says MP opens 44+, A2s+, K7s+, Q9s+, J9s+, T8s+, 98s, 87s, ATo+, KTo+, QTo+, JTo and BTN calls TT-55, AQs-AJs, KJs-KTs, QTs+, J9s+, T9s, 76s, 65s, 54s, AQo, and that's what you want me to try? Presumably Snowie opens 2.5bb, compared to Janda's 3.5bb. Also, I expect that rake will have an effect on all these calculations, and I think Snowie tries to take that into account. I can add rake to my calculations easily enough, but I haven't tried it yet. Can you see whether Snowie gives different results if you make it NL10 compared to NL1K?
I changed stakes to 20kNL so ~rake free. MP oR to 2.25bb 44+, A2s+, K7s+, Q9s+, J9s+, T8s+, 98s, 87s, 76s, ATo+, KTo+, QJo, JTo. BU call 99-22, TdTh, TdTs, ThTs, AQs, KJs-K9s, Q9s+, J9s+, T8s+, 98s, 87s, 75s+, 64s+, 53s+, AdJd, AhJh, As9s, Ac8c, AQo. So yes different ranges then on NL200. And MP cbet 1/2 pot on 886r BU has no raising range. It would be very interesting compare solutions.
Originally posted by dooleslovs
I changed stakes to 20kNL so ~rake free. MP oR to 2.25bb 44+, A2s+, K7s+, Q9s+, J9s+, T8s+, 98s, 87s, 76s, ATo+, KTo+, QJo, JTo. BU call 99-22, TdTh, TdTs, ThTs, AQs, KJs-K9s, Q9s+, J9s+, T8s+, 98s, 87s, 75s+, 64s+, 53s+, AdJd, AhJh, As9s, Ac8c, AQo. So yes different ranges then on NL200. And MP cbet 1/2 pot on 886r BU has no raising range. It would be very interesting compare solutions.
OK. It'll certainly be challenging. The wider the ranges and the deeper the stacks, the longer the computation time. I think I need to get the code parallelized before I try that one, but thanks for your interest in my project. I'll keep you posted, and I'll put 'add rake' on my To Do list.
OK thanks it is very interesting I downloaded your program was hoping to put ranges and boards myself ♦ But it seems it's long way until that stage.
Originally posted by dooleslovs
OK thanks it is very interesting I downloaded your program was hoping to put ranges and boards myself ♦ But it seems it's long way until that stage.
There are still lots of technical issues to deal with (working on it part time since July is not long), but I'm happy to do requests, and what you've asked for is interesting. If I ever do release a version that lets you run the algorithm yourself, you'll need a decent PC with plenty of RAM. I've been running example hand 1 with three bet sizes for a couple of days now, and it uses about 3.5Gb (my laptop has 2Gb and fell down dead on its back with its little legs pointing in the air when I tried to run it on that by mistake). The solution will probably be ready by tomorrow. Exactly how many iterations are needed for the solution to be informative is a good question, and a different question to 'How many iterations do I need before the solution is close to unexploitable?', which is what most of the literature tries to address. I suspect the answer is probably 'less than I'm doing at the moment', but we'll see. I'm writing a learning tool, not a bot. Any feedback you can give me on what you can learn from these solutions, like Jules's excellent posts, would be very helpful.
Originally posted by jbpatzer
I get exactly 50.5% from Equilab, so I think you must be entering the ranges wrong.
yeah not surprising,
if it's very easy it would be nice if JBR spat out the preflop ranges somewhere, preferably in an exportable way that pokerstove/equilab/crev could read.
I've now modified the code to be able to do turn and river situations as well. I just need to do a bit more work tonight, and I'll test it on some obvious cases, and also run it on GTO Brainteaser #8 in the GTORangeBuilder blog so that I can post the solution there like the smug, obnoxious bastard that I am.
haha, yeah thats a good plan ♦
I had thought of giving it a crack manually, but that seems like too much work.
fwiw, my guess is that, the only difference between this and the previous example, is on the river when an, I think it was a 2c or 3c comes, whatever blocks hero's bluffs. So villain can alwyas fold his KK. An A on the river changes nothing. So slightly less EV than the previous brainteaser. And my guess was, that the betsizing on the turn slightly increases relative to the previous example to make up for this, to create a cost for the information received seeing a river.
If the betsizing changes, than maybe your smug, obnoxious solution might not pick it up right? haha
Originally posted by jules97
Originally posted by jbpatzer
I get exactly 50.5% from Equilab, so I think you must be entering the ranges wrong.
yeah not surprising,
if it's very easy it would be nice if JBR spat out the preflop ranges somewhere, preferably in an exportable way that pokerstove/equilab/crev could read.I've now modified the code to be able to do turn and river situations as well. I just need to do a bit more work tonight, and I'll test it on some obvious cases, and also run it on GTO Brainteaser #8 in the GTORangeBuilder blog so that I can post the solution there like the smug, obnoxious bastard that I am.
haha, yeah thats a good plan ♦
I had thought of giving it a crack manually, but that seems like too much work.fwiw, my guess is that, the only difference between this and the previous example, is on the river when an, I think it was a 2c or 3c comes, whatever blocks hero's bluffs. So villain can alwyas fold his KK. An A on the river changes nothing. So slightly less EV than the previous brainteaser. And my guess was, that the betsizing on the turn slightly increases relative to the previous example to make up for this, to create a cost for the information received seeing a river.
If the betsizing changes, than maybe your smug, obnoxious solution might not pick it up right? haha
A on the river does change things, because player one can't bluff his 32o anymore.
I can try various bet sizes to see what's most +EV, but no, I'm less smug about that, although probably equally as obnoxious.
My understanding is that pokersnowie is the result of a neural network. I'd be surprised if the solutions were identical. If they are... than I guess I'll be subscribing to pokersnowie again. It would be interesting to see the EV differences.
Originally posted by jbpatzer
A on the river does change things, because player one can't bluff his 32o anymore.
I can try various bet sizes to see what's most +EV, but no, I'm less smug about that, although probably equally as obnoxious.
I thought all an Ace should do, is change hero's 0 ev bluffs to 0 ev folds.
Originally posted by jules97
My understanding is that pokersnowie is the result of a neural network. I'd be surprised if the solutions were identical. If they are... than I guess I'll be subscribing to pokersnowie again. It would be interesting to see the EV differences.
For a given flop and given preflop ranges, if PokerSnowie's neural network converges to anything other than a GTO solution, it's not doing its job, given that it's trained by playing against itself.
Originally posted by jules97
Originally posted by jbpatzer
A on the river does change things, because player one can't bluff his 32o anymore.
I can try various bet sizes to see what's most +EV, but no, I'm less smug about that, although probably equally as obnoxious.
I thought all an Ace should do, is change hero's 0 ev bluffs to 0 ev folds.
Hmmm. Not sure, but I would have thought that the possibility of an A appearing on the river would make bluff more on the turn.
However, the GTO solution I calculate has him bluffing slightly less on the turn and more on the river, so wtf do I know. I posted what I found on the GTORangeBuilder blog.
Apparently I got it right. Well done me! ♦ But seriously, it's nice to have an independent check that my code is giving sensible answers in simple cases.
Yeah, I'm skeptical pokersnowie is doing it's job as advertised.
Nice job with the brainteaser solution. Didn't expect anything less. ♠_biggrin:
The solution has me confused. 
The turn I can kinda get. Hero's range isn't as nutsy (don't laugh), so he can't bet like if it was purely polarized.
But on the river, betting more air?!
I think I'm going to have to wait for someone to walk through this for me. It looks like it may have implications that could come in handy in the plethora of spots where new cards alter hand strength. In this example the change is small, but others it's massive.
Anyways, I don't get it at the moment...
I had a look at the recent 3 betsizes A:d7:s5:d and compared it to the earlier single bet size one. I'm sure you've done this already.
Both were checking 88% and betting 12%. This seems a little odd? But not completely out there, so yeah, I don't know. I take it that 'indexing routine bug' you mentioned is insignificant? I just kind of thought before I looked at it, that these frequencies would change a little.
So, the only difference between the two is bet sizing. When given the option it bets smaller more often. Which to me makes sense intuitively, ranges mostly suck so betting smaller should eek out a little more value.
The bit I found most interesting was the difference in EV's. 3.76bb/4.76bb with the 3 bet sizes, verse the 3.74bb/4.76bb with the one betsize.
So what I want 
to take away from this is that betsizes don't matter too much. I'm happy to give up .5% of winrate to forgo a 1000% increase in effort.
