Comment by mattmaroon | Hacker News Reader

Comment by mattmaroon | Hacker News Reader

Parent Full thread

I don't think I agree with you on no-limit vs. limit. There's a pretty simple strategy you can use to be a marginal winner at low stakes no-limit (or especially no-limit sit and goes). It' called short stacking, and essentially involves buying in for the minimum and playing mainly all-in or fold (with some preflop calling).

You have to leave the table as soon as you get past 1.5x the minimum buy-in though. It was mainly done on Party Poker, which was the only site dumb enough to let you cash out and then return to the table for the min again. Any other site either made you bring back what you left with or wait an hour.

On a site with enough tables in action, though, even that would not be a limitation. A bot could do fairly well at it.

Also I can tell you from first hand experience that breaking even at low stakes NL sit and goes can be done with an AI consisting of nothing but if/then statments. I did this with WinHoldem back in the days before it was banned from everywhere. I wasn't able to replicate that success with limit. The main problem with WinHoldem was that it didn't join or leave tables on its own (also, it would often bug out when the tournament got heads up), which meant that the time spent managing it made it less profitable than just playing myself.

That's why we tried to build a bot ourselves. Maybe I'll blog about that some time.

Yes, I looked up a few poker bots yesterday and those sites said that the bots are generally more profitable on no-limit with experienced (predictable) players. I can dig up the links again if anyone wants.

The reason limit is easier is just that there is less variables to account for much like it's easier to write a bot to play poker with fewer cards.

While I will agree sit and gos are all strategy that strategy is dependent on putting your opponent on reasonable ranges which a bot like this won't do.

Bots and winning bots are possible(stick most casual people up against the UofA bots for example), but I find it hard to believe this one is one of them.

That's wrong too. If you think there are fewer variables in limit hold'em to figure when making your decision, you don't understand the game.

If you went to a no-limit tournament and just pushed all of your chips into the middle at every single opportunity, it's been shown that you're only a slight loser. At the WSOP main event, you'd be a better player than at least 1/3 of the field.

This is not true if you were to play in a limit tournament and raise every time. You'd probably be the worst player in the field.

The fact that it's less complicated is why no-limit is the tournament game of choice, but high stakes live poker is generally limit.

There are fewer variables when factoring in your opponent modeling. I'm sure being who you are you are very familiar with the UofA research and I'm pretty sure it backs up this statement.

Pushing every hand isn't right, it's pushing every hand with certain cards that Sklansky taught which admittedly is only a slight loser but a loser none the less.

I'm not saying limit holdem isn't a complex deep game, all I'm saying is that when it comes to bots no-limit bots are a lot harder to get right(in opponent modelling primarily which is a big part of bots these days).

I think no-limit is the tournament game of choice because it's seen by the general public as being the more exciting of the two(not that I agree but thats another discussion).

I still disagree and think that limit is, in general, a much more complex game. But even if it's not, I think you're thinking of building a bot that's a world class player. That may be harder at no-limit, I'm not really sure.

But if you're building from the standpoint of breaking even at low stakes, it's far easier to do for no-limit, especially sit and goes. It's literally a couple hundred lines of if/then statements.

Nope. NL is a much more complex game than limit. This is covered fairly well in the U-of-A literature. Also some of the two plus two NL books describe why NL is more complex.

Just because there is a simple strategy (shortstacking push/fold) that can be applied successfully against weak players, that doesn't mean that NL is a simpler game.

When you are trying to model the game there is a lot more to consider in NL. In Limit at each point in a hand and at each future point in a hand there are only limited actions that your opponent can take - raise, call, check, fold. In NL they can do any of these actions but also for any amount - push, raise pot, raise 1/2 pot etc. So the problem space for NL is much larger than for Limit. Bluffing is a much bigger part of NL and bluffing is hard to model.

Playing profitable limit poker is a fairly mechanical, mathematical process. You can't force people out of a hand with a big bet. It basically boils down to betting when you're probably ahead, folding when you're probably behind and drawing when the pot odds are favorable.

The U of A bots can play limit reasonably successfully against decent players. A decent player can easily beat their bots at NL.

Building a bot that can profit at the microstakes tables might be easier for NL than limit because you can employ a push/fold short stacking strategy like you describe. But building a bot that can survive against better players would be much harder for NL than Limit.

Reply on news.ycombinator.com