Midlife Crisis? What Midlife Crisis?

But can they play?

Originally posted by EmanuelC16
But can they play?

Apparently not.

Originally posted by jbpatzer
I just listened to

http://www.thinkingpoker.net/2014/05/episode-79-the-computer-poker-research-group

Really interesting interview with two computer poker chaps from Alberta. Confirms everything I thought about multiplayer games. Well worth listening to.

Cool topic.
Seems they are of the same opinion as you with multiplayer games and collusion. It's interesting because to me it sounds counterintuitive regarding the effect that a weaker player at the table has. I take away that I'm supposed to play more like a monkey in this situation.

When are you building your AI poker bot for multiplayer games? Any idea where to start about doing this? Considering all the AI that is made only are really effective in 2 player games and when adapted go multiplayer games kind of suck.

I can see a neural network along the lines of poker snowie being ok, but they would have to start from scratch and run a few different simulations with say 1 fish consisting of a few different types of 'fish' at the table. But surely there's a faster way to get there with some logic.

Originally posted by jules97

When are you building your AI poker bot for multiplayer games? Any idea where to start about doing this? Considering all the AI that is made only are really effective in 2 player games and when adapted go multiplayer games kind of suck.

I can see a neural network along the lines of poker snowie being ok, but they would have to start from scratch and run a few different simulations with say 1 fish consisting of a few different types of 'fish' at the table. But surely there's a faster way to get there with some logic.

I don't think I'm going to be botting my way to glory any time soon. It's really not my speciality. I'm more of an 'abstract the essence of a problem and put it into a simple model to gain insight' kind of a guy.

In my opinion, the reason that multiplayer games are hard is that you have firstly to develop a model of you opponent's strategy and secondly to work out how to maximally exploit them whilst thirdly being alert to shifts in their strategies. Neural networks like PokerSnowie are, as far as I know, static, in the sense that they are trained to play a certain strategy, whatever the opponents are doing, and are therefore fundamentally incapable of adjusting to the composition and actions of the table. Although PokerSnowie is evolving over time, its on a time scale of millions of hands, not the time scale needed to react to a table.

Opponent modelling is hard. In the book, I refer to a paper where the authors studied basically the AKQ game, and allowed their bots to play various dynamic strategies. The results were not entirely trivial. Even though the optimal solution is trivial in this game, imagine sitting playing against someone who never bluffs with his Queens. It would take an iron will to keep calling with 1/3 of your Kings against someone like that. You'd probably start folding them all. Then your opponent could start bluffing...and we're off down that route again. In multiplayer games, where there are likely to be multiple equilibria, the problem's even worse.

I'm confident that 6max NLHE isn't going to be 'solved' any time soon, and the people who make the most money are going to be those who can think outside the box (i.e. play close to an equilibrium strategy that's not the usual one, e.g. limping some buttons and SBs, unusual bet sizings, calling 4bets), work out their opponent's strategies quickly and start exploiting them, and be alert to shifts in the dynamics of the game. Unfortunately, that's probably not going to be me. Fml, poker is hard.

Originally posted by jbpatzer
I don't think I'm going to be botting my way to glory any time soon. It's really not my speciality. I'm more of an 'abstract the essence of a problem and put it into a simple model to gain insight' kind of a guy.

Screw that attitude. It's not just poker bots, it's Terminator II Judgement Day AI. And if you don't do it, google will.
The fate of humanity is in your hands.

I was working on a BTN limping strategy recently. I figured if it works from the SB, it should work from the BTN too! It got tricky having to consider the SB. I think it could work though. He gets great odds but is sandwiched so my theory was that he can't really play that much anyway.
In the end I sort of figured I could play a few more hands from the BTN if I limped, but most aren't playing well enough from the blinds to make it worth while anyway and I can already open those hands very often. So my efforts are probably best spent elsewhere. Such as not calling river donk overbets from fish.

Originally posted by jules97
I was working on a BTN limping strategy recently. I figured if it works from the SB, it should work from the BTN too! It got tricky having to consider the SB. I think it could work though. He gets great odds but is sandwiched so my theory was that he can't really play that much anyway.

I see a LOT of players who limp BTN in the head-up part of SNG, then fold to BB raise.

I don't see many limping BTNs in cash games.

If only the one limper, is seems an ideal time for a x/r from the BB OTF -- no?

--VS

Yeah I was looking at cash games 100bb or a little deeper, I have no idea what the guys you are seeing in SnG's are doing limping.

With a massive SPR, position & some solid ranges I'd spent some time on (and I assume most BB's haven't) I think seeing raises often would actually be pretty good thing.

My BTN defending raise range was something close to a normal BTN opening range. That was the fun part of working on this, making these super wide defending ranges on the BTN. Getting into all these preflop 3b/4b/5b/etc games ♦.

The problem with a x/r as BB after preflop goes limp/limp is that BB's range is just trash. Unless he flops a very rare flop of something like 235 he's just about the biggest dog one can get range v range when we see a flop in poker. So he has bugger all value hands to x/r with and even when he does, they are often dominated by BTN's range. And with the SPR so deep hero can really put the pressure on.

So I wouldn't think x/r after limp/limp would be the best idea. I think unfortunately for the BB being OOP he just has to play well OOP and deal with getting put in some bad spots. On the good side for BB he will get to see a tonne more flops and realize more equity. But he's going to have to work hard, play well postflop and deal with some horrible spots to realize this equity.

I've spent a lot of this rather damp Bank Holiday weekend reading

which is meant to be the best and most comprehensive text book available on Game Theory. It's the size of a telephone directory, and costs an arm and a leg, so it's lucky that I could get the University to buy it for me ♠_cool: . It has lived up to its reputation, and I feel I start to see how multiplayer game theory works now.

Here's something that rather took me aback. Let's imagine that you're playing a nonzero sum game with some hidden information (like poker for example). I would have thought that giving one of the players access to some of the hidden information would increase his payoff. Wrong. 'Wtf!' I hear you shout. Here's a simple example.

Consider the following game. Player one can choose either to play rock/paper/scissors for $10 or take a guaranteed win of $1 for both himself and player two. If he chooses rock/paper/scissors, the optimal strategy for each player is to play one of the three options at random, which gives them both an equal expectation of $5. Player one should therefore choose to play rock/paper/scissors since he makes more than the guaranteed $1.

Now let's give player two some information (Weeee! More information = win, right?). If player one chooses to play rock/paper/scissors, player two gets to see what player one plays first. If he plays rock, he can play paper, etc, so player two is guaranteed to win $10, and player one gets nothing. Splendid! However, player one now prefers not to play rock/paper/scissors, and instead chooses the guaranteed win of $1 each instead.

Original game - both players win $5. Original game + more information for player two - both players win $1.

It seems like a paradox, but there is a resolution. See if you can get your head round it before you look in the spoiler.

oneday you may even solve multiplayer chess.. oh wait

Originally posted by mineriva
oneday you may even solve multiplayer chess.. oh wait

Now my head hurts again.

Originally posted by gadget51
Now my head hurts again.

Because it's been set up with the King and Queen the wrong way round?

Ok I take white .. e4 (which is now Queen pawn)

Originally posted by mineriva
Ok I take white .. e4 (which is now Queen pawn)

Mal and I are going to collude against you, so you might as well resign and let me play him HU.

Originally posted by jbpatzer

Originally posted by mineriva
Ok I take white .. e4 (which is now Queen pawn)

Mal and I are going to collude against you, so you might as well resign and let me play him HU.

No you have not solved the game yet and even if you did i think Mal understands just as must of your scrabbles as I do. Which is very little.

So I am safe for now. ♦

PS> She who must be obeyed will not allow him to collude. ♠_biggrin:

Originally posted by mineriva

Originally posted by jbpatzer

Originally posted by mineriva
Ok I take white .. e4 (which is now Queen pawn)

Mal and I are going to collude against you, so you might as well resign and let me play him HU.

No you have not solved the game yet and even if you did i think Mal understands just as must of your scrabbles as I do. Which is very little.

So I am safe for now. ♦

PS> She who must be obeyed will not allow him to collude. ♠_biggrin:

I think if we both decided to take your pieces but not each others, we'd do just fine. I keep saying that poker is infinitely harder than chess, and that goes for multiplayer chess too.

Ooooooo! Ok then.

Challenge accepted?

Originally posted by gadget51
Challenge accepted?

Yep. I already feel that switching to PT4, with its splendid automatic note taking system, has made a big difference to me, and I've started playing some NL50 again. I may even show a graph at the end of the month.

I just finished my MTT night. I fired up some 180 mans, and got to three final tables, which I've never managed before. Very confusing, particularly when my mouse ran out of batteries half way through. I managed 2nd, 3rd and 4th, which was kind of annoying, but more profitable than one 1st, so whatever.

Here's my Zoom graph for the month, mainly NL25.

I felt like I played a lot, but it's still pretty pathetic volume considering what I used to manage, and I'm going to be away a lot next month. I have five days teaching at a Summer School in Nice for PhD students on an EU project, and my final week in Mauritius before they have to look for a new external examiner. Maybe I'll try the casino again....

I can also report that I've managed to persuade the nice man from the University of Alberta who was interviewed on the Thinking Poker Podcast to exchange emails with me about poker and game theory. I told him that my poker friends aren't mathematicians and my maths friends aren't poker players so I had nobody to talk to, and I think he must have felt sorry for me.

If that all sounds good, I should point out that I have 260 first year exam scripts to mark next week and a day at UCL looking through their exam papers on Friday, so fml after all.