SwissDollars
77 points
Hi guys,
There was a thread discussing Pluribus playing style here but nothing on what it means for online poker, so I thought I would start one here so we can each share our thoughts.
Here are a couple of interesting quotes from Noam Brown, the co-creator, which answered a AMA 5 days ago
We want to make the research accessible to AI researchers, so we're
including detailed descriptions of the algorithms and pseudocode in
the supplementary material, but we won't be releasing the code or
models in part because it would have a serious impact on online poker.
....
Pluribus, by comparison, uses "$150 worth of compute and runs in real
time on 2 CPUs" using less than 128 GB of memory
....
The most popular poker sites have advanced bot-detection techniques,
so trying to run a bot online is probably too risky to be worth it.
But I do think this kind of research will have an impact on pro poker.
In particular I think our latest techniques will be adopted by poker
training tools. Those tools are particularly weak right now when
dealing with 3+ player situations. Things like Linear CFR and
Discounted CFR should also allow these tools to compute all solutions
faster than they currently do.
....
Pluribus does not adapt to the way its opponents play. It treated each
hand that it played against the humans individually and did not carry
over knowledge from one hand to another. It learned to play entirely
through self play.
....
CFR is not guaranteed to converge to a Nash equilibrium in bridge.
That said, it wasn’t guaranteed to converge to anything useful in
6-player poker either, but it worked fine there.
....
I think opponent adaptation/exploitation is still a very interesting
AI challenge. I do think that top pros could beat weak players by more
than Pluribus would (though I do think Pluribus would still make a ton
of money off of weak players). The current state of the art for
opponent adaptation is pretty disappointing.
Tough the code for the bot is not freely available, it is partly available to other AI researchers, so there is a good possibility that it will get to some people with the right contacts and be developed to play on the larger sites. Fraud is always ahead of the anti-fraud detection by 6-12 months, as we saw with the Oborra or some Eastern European PS accounts. I could also imagine a real-time advisor being developed out of Pluribus play since it's really fast and only worth $150 of compute. All this would impact the poker online ecosystem. Are the days of online cash game counted? And how would you adjust your study habits in the short-/medium-term?
July 24, 2019 | 3:15 p.m.
good session Chris
8:07 is 88 here 16.5bb eff SB vs BB not strong enough to induce raise/call?
Jan. 16, 2019 | 12:47 p.m.
good stuff right here, like your in-game analyses
11:15 not a thing to raise BB vs SB limper with 54o here?
14:15 can we find a better spot than putting 38bb with AJo BB vs BU even though it's an HS tourney? or is it just standard since we can't 3-bet/call and just call is weak with AJ vs BU wide open?
Jan. 16, 2019 | 1:03 a.m.
Great stuff, I like your comment on PIO adjusting villain's strategy on future streets due to us node-locking his flop strategy and being mindful about it. Also how much those strategy adjustments are really worth in EV to us. Like Nihithaw, I'd have appreciated node-locking the OOP over c-betting and also looking at a single-raised pot vs BB and which hands we 3-barell-bluff. keep up the good work!
Jan. 15, 2019 | 11:21 p.m.
Hi Juan, great useful video, thanks a lot!
5:10 Flop report OOP
Why do PIO prefer as the highest frequency option (40-50%) to CHECK OOP on those 2BW + rag type of boards with such high equity & EV advantage?
Kd Qs 2s - check; 58% equity
Ad Js 3s - check; 58.9% equity
Ad Ks 7s - check; 58.4% equity
As Qs 7s - check; 57.7% equity
Jan. 9, 2019 | 11:18 a.m.
hi Mark, your exploit vs a OOP 100% c-bet on most neutral / better for IP boards was to raise IP more often, which I totally agree with and is useful since most players in my live pool c-bet way too high on almost any boards and never check/raise as PFA (unless with sets). Tough we often lack credibility when we do it, they still can't defend enough anyway
however when I was running sims (with more loose range for the opener like HJ/CO or vs BU cc), to my surprise, on some board like J43r, the exploit was not to raise IP more often but to call with 100% range (the problem is when we call 100% and they keep barreling on turn, our IP continuing range is so weak)
how can you explain that on some boards the raise frequency stays the same vs over c-bet and the right PIO strategy is to mostly call and not raise? how do you then proceed on later streets ? Thanks!
Jan. 7, 2019 | 8:53 p.m.
very insightful video! i like your comments on weaker players / low-stakes regs thinking as well as the fragility of equilibrium when dealing with narrow ranges
on the last hand, I presume this would impact OOP turn leading strategy after we raise the flop IP with JJ? if OOP is not 3-betting flop enough, he must then now lead turn after calling our raise, do you need to lock that as well to get more accurate result?
Jan. 4, 2019 | 3:42 p.m.
good short video Chris, 4-tabling is good
At 4:35 I agree with the A2o fold pre-flop vs bigger sizing vs ONE opponent description here. running into equilab this is a fold whereas A3o is a call.
As we're multi-way and we now get better odds, shall we extend our calling range or similar to PLO call tighter in BB multi-way and fold those hands which play bad multi & have reverse implied odds like here? what if the two opponents are weak bad, this become a +EV call i reckon?
8:58 is ATs BB vs BU not strong enough to be part of our 3-bet value range? I generally 3-bet 50bb+ BB vs BU 99+, AJo+, ATs+, KQo, KJs+ which allows me to have more 3-bet-bluff in that spot too
Feb. 26, 2018 | 1:59 a.m.
40:30 what about installing a BU limping strategy in Tom's shoes given ICM consideration? i think this is what i'd have preferred to keep pot small and not have to raise/fold A7s
50:00 should Fedor not use a raise-only strategy vs Tom due to the presence of short stack and how tight Tom must call PF and cautious post?
Great video. I'd be great if you could cut the in-between hands and non-interesting PF hands so they're less dead time. otherwise top content and analysis
Feb. 17, 2018 | 1:10 a.m.
thanks for the precision shallow vs deep stack, very useful
Feb. 13, 2018 | 5:45 p.m.
16:30 is 63ss on J54cc it not a combo we'd prefer to check back as we can't bet/call GII here and prefer like GS or BDFD+BDSD to bet/fold?
46:24 i think PIO want us to bet in general hand with higher equity, e.g 2nd pair top kicker with BDFD+BDSD here and check back AThh with no BDFD. it might be counter-intuitive as this hand seem great check back having the bckdr covered
Feb. 13, 2018 | 1:42 p.m.
quick ICM pre-bubble question:
* we are big chip leader and cover the 2nd biggest stack by well 25-30%. Why are we so concerned of playing low variance vs him when we can build big pot, apply ICM pressure post-flop and being able to have more bluffs vs him?
* his risk premium vs us is much higher than us vs him and is also much higher than the short stacks risk premium vs us. If we were shorter than him, it's ofc makes sense to play pot-control line close to the bubble, but aren't we too cautious vs him given he is ICM-handcuffed and can only fight back with strong hands?
Feb. 13, 2018 | 9:16 a.m.
i really appreciated the video, high-quality content from one of the best performer ITW, awesome thanks!
15:00 OTR it doesn't seem we have many bluffs here calling flop 3-way and our Qx/Kx in spades have now SD value and our JTs now is a straight. J9s, T9s, 98s, 87s are the only naked busted flush draws with no SD value to bluff with. our value range is pbbly very narrow too KQ/AJ. Most of villain's range we're are targeting are Kx, Qx, weaker Ax, pocket pairs. Given the above, it seems more GTO to bet in the 30-50% range. any thoughts?
Feb. 12, 2018 | 11:06 p.m.
always a pleasure to watch your videos Sam! i was at PS Bucharest last year when you won the main event but failed to say hello! you seemed in the zone, did not want to disturb, hope to connect on another poker stop!
07:30 i think it especially plays better as a XC as we're at FT vs the only stack who covers us, which should incentivize us to use pot control line. not a fan of the raise with this combo
Feb. 11, 2018 | 2:09 a.m.
hey grayson, good short turbo video, thanks for it. a few comments/questions on my side:
22.50 i don't think we can raise/call ATo vs BU given risk premium. raise/call range should look more like 6.0%, 88+ AJs+ AQo+ whereas AJo is break-even
Feb. 10, 2018 | 11:49 a.m.
Great series, a lot of cool insights, thanks Chris! 2 questions:
16:37 I inputed this spot in HRC; given BU is so loose and can open so wide being the big stack, SB can reshove 24.2%, 22+ A2s+ A8o+ K7s+ KTo+ Q8s+ QJo J9s+ T8s+ 98s. So 98s would have been pretty good even given ICM considerations
30:22 given ICM here in villain's shoes and a strong suited hand i think open-jamming tend to do better than min-raising. At least this is what HRC indicates. Even jamming AKo would be better here as mid-stack with 2 short stack. Only TT+ would do better as min-raising. What do you think about open-jamming in that spot?
April 14, 2017 | 3:16 p.m.
Hey Chris, regarding your reply above:
* ICM says that your AJo 3-bet-shove makes ~+0.05% EV of prizepool left so +$11'800 EV shove vs a 17.4% HJ open. A pretty good play given he must call so tight (ICM says 2.1%, JJ+ AKs only) so not a mistake!?
April 14, 2017 | 3:10 p.m.
Great new content Sam! Loving the thorough use of third-party software to back up your play, very professional approach. 2 questions:
6:30 you went over your 3-bet range BU vs CO ~25bb eff. Can you detail how you come up with the 2 to 1 value to bluff combos as the optimal ratio in that spot? And when would you deviate from that ratio?
38:00 How does your 3-betting range look in that exact spot? would you also use a 2 to 1 ratio here? he is opening from a 27bb stack in CU and we're in BB
April 8, 2017 | 9:01 p.m.
I really loved the video, loved the topic, great work in an area where so many people misplayed; i'd absolutely love to see more on the topic be it your suggestion above Shak vs Selbst hand / or analysis of some FT chip distribution and how it affects every one
Just one comment on my side, at 42:30 you said that with less chips (300k)) we should fold even more. I agree that having less chips 300k (3bb) does not change our probability of winning a potential heads-up match should we call and win. But folding actually decreases substantially our chance of finishing 2nd and if 3rd place then doubles up next hand, we will be in tough place so we should loose up just a little bit our calling range to guarantee 2nd prize here and put some distance with the 3rd place stack. With more distance to 3rd place stack, then it becomes right again to fold tighter
March 29, 2017 | 1:06 a.m.
Great series so far Dylan, thanks and i really appreciate the max. 4-table format, 6-table was way too small
02:00 i was curious about this 20bb eff. spot & run it into PIO and it says that we should check back our 77 here on the turn, for villain to check twice, he really has more 2nd/3rd pairs than nothing combos and can still have some turned top pair. his nothing combos with equity would have lead turn so those that check generally are dead or only 3 outs and we don't need to protect really
04:30 you raise-folded here AJo when covering the reshoving stack with a big $150 bounty and i'm not really sure if that is the correct play?
- as we pay $25 and get 5k start stack, so $75 bounty should be worth 15k, but because we only win half of the bounty directly, its approx. direct chip value should be ~7.5k is that more or less correct?
- so here in this spot, villain has $150 bounty so it's extra 15k in pot for 31.6k total and we need to call 12k given 2.6 to 1 odds? but shall we loose we give up too much value in terms of exploiting the bounty prize pool later in the tourney so you prefer a fold here for that reason?
- if we fold here, i think we are better off just open shoving those hands we want to play at this stack depth in a SPKO?
Interesting to get yur thoughts on this spot
March 28, 2017 | 3:45 p.m.
Very tedious work you did there, thanks a lot i re-wrote most of it in my notes
One question at 39:00
Q![A, K 7] has 55% equity whereas KQ![A,7] has 53% equity, what is the reason for it? why KQ![A,7] which should be ahead of Q![A, K 7] as this combo must hit 3 live kickers, has lower equity due to having a king?
Just a tip, when i re-wrote all this in my notes, i actually use the cards to illustrate the flop and helped me a lot just to get the material in. You can just google "poker cards svg" and use those in your presentation would make it more human :)
I look forward to the next vides in the series Core PLO Concepts!
March 27, 2017 | 11:52 p.m.
Sam, i think you did not include the OOP bet option at 6:00 thus PIO indicating checking 100% of times. Cheers
March 24, 2017 | 11:48 p.m.
Hey Sam, awesome video, i just got the elite sub again and this video is at least 10x worth it! I've used the same polarized river bet strategy in the past and this gives me new strategy to explore with PIO for my 50-60%ish equity combos OTR. Thanks & Cheers!
Just one comment, how does this apply in live setting? I've experienced that a lot of villains don't value bet thinly enough OTR to put us in tricky spot with our bluff-catchers, Are we small value-owning ourself with the very bottom blocking bet hands with 50-55ish equity when we can actually showdown those hands for free or exploitatively check-fold vs a river bet? on the other hand, we might get called by Ace high so I see some merits for it even vs those type of villains
March 22, 2017 | 7:43 p.m.
i enjoyed this video, thans gray! i know the video is already 4-month old but would appreciate if you find the time to answer y couple of questions
01:00 i like your HR line but I was wondering why you decided to deviate so drastically from PIO solution by checking back 5 high on this flop?
our opponent will not be able to defend 1-@ on this flop and its' a great hand that can barrell many turn/rivers
10:45 i think we have such a range advantage on this board as villain would have 4-bet shoved all his strong Ax that we should be betting here for a small amount with KQ.KQ with diamond can be a good candidate to check back to not be blown out of our equity by a XRAI from villain. You play your particular hand by checking back here but PIO also indicates 85% betting frequency for a small $1-1.1k, incl. most KQ
Jan. 12, 2017 | 6:40 p.m.
Stellar content here, well done Chriss, I definitely like to see more of similar type videos
Very interesting to see that our edge is actually the biggest at low 25bbish stack depth instead of 45bb/85bbish stack depth
a) did you allow in your flop simulation for turn/river play? can you share the files somewhere?
b) have you practically implemented in your own game some check-raising with strong top pair Tx at 85bb stack depth vs a UTG range?
Jan. 11, 2017 | 6:34 p.m.
This is an amazing in-depth video, thanks Daniel for crafting it!
question on a little detail how you constructed your flop betting range:
why did you have the 77-66, 44-33 as betting with a diamond but checking with no diamond? I was thinking the one without diamond needs more protection and need to bet whereas the one with diamond can play better as XC less fearing a diamond turn
Jan. 11, 2017 | 1:07 p.m.
#1 Why do you take most of the profitable shoves with off-suit combos & construct a raise/call and raise/fold range around suited combos?
very interesting coz I actually do almost the opposite. 2 reasons:
a) most villain's re-shoving ranges are much more weighted towards suited hands instead of offsuit hands; by raising a suited combo (e.g A5ss), there are 3 suited combos of the same hand villain can re-shove. By raising an offsuit combo /e.g A5o); there are only 2 suited combos of the same hand, so we actually decrease the probability of being re-shoved on
b) most suited combos are too good to raise-fold but not good enough really to raise-call, so i preferred open-shoving them, when called we have a small equity boost as we're suited; This is based on analyzing HRC open-shoving bb/100 with an open-raise bb/100 and an open-raise-call bb/100 and see which hands fit best where in our range
#2 Great point Sam, if we have a very solid opponent at shallow stack depth, it will be easy for him to defend & play correctly post-flop. By open-shoving we remove him all his options and take the bb/100 EV
Jan. 6, 2017 | 10:37 a.m.
good to see a video focused on push/fold , push/call, good analysis, thanks Sam
A few comments:
#1 2:50: what are your stats here: VPIP/PFR/3-bet? if so BU has 2% 3-bet and SB never 3-bet shove so far, I'd prefer an exploitative raise-fold for those reasons; also we block Ax increasing our steal success %
#2 8:10 I actually disagree, compare with HRC open-shoving bb/100 with a simple open-raise bb/100 and a open-raise-call bb/100, you'll see i think that A9o generate the most bb/100 as a raise-call. However if there is any ICM consideration it might become an open-shove as we now must call tighter
Jan. 3, 2017 | 3:02 p.m.
great video Sam, always insightful stuff
12:30 another disadvantage of flatting here is our flush is only 8 high and we might run into better flush. By raising, we eliminate many better suited hands K2cc, Q3cc that we'd loose a lot of chips to, should the flush comes in. You did iso 65cc a few hands later for those reasons so not isolating 87cc because we've been isolating a lot don't seem to be a valuable reason, unless villain has shown proportion to fight back or opponents behind have taken advantage of it, which hasn't been the case so far. what do you think?
small tip for you Chris, use "SHIFT + RIGHT ARROW" to go the next hand VPIP in PT4, i hope it will help your video production :)
9:40 why do you use a 60% bet sizing? it seems we don't have many bluffs with that line OTR
Jan. 19, 2019 | 12:14 a.m.