No fear!

The code to solve the game is open. You can go download it right now, and run it on your PC with a pile of 6TB drives. Have fun, see you in ten years from now (unless you've got your own research cluster - that IS possible...) It's always been a question of available resources. If you ignore practicality, you could have solved poker back in the 1950s. Koller's game theory work in the late 1990s made it interesting. 2007 and CFR made solving limit Hold'em seem out of reach, but not crazy. 2014 just sped things up.

As for being worried about someone breaking in and stealing the 11TB table of numbers that lists Cepheus' strategy. Really? Can't tell if serious...

totally serious, but I'm not a tech savvy, so you know, no knowledge=stupid questions.

Ty for your time for answering

Okay, then have no fears about someone stealing Cepheus: even if you walked in with your own hard drive, and did a copy command, it would take more than a day. It's a BIG file.

They steal the code... then they just need to find 4800 cores and a lot of disc space to run it for 2months. Then they need to find someone to play against in HU LHE and then they need to find a way to get assistance from their newly calculated solution while playing.
I mean... there are lower hanging fruits out there if you have this kind of resources and want to engineer a scam.

Link please ? (Cepheus home page is unavailable right now)

Wow, i hope this is an A* troll. It is quite brilliant.

Let me dumb this down, because a lot of people are dumb like me but I think the gist of this is:

On this website http://poker.srv.ualberta.ca/strategy# (down right now) you can literally in real time ask it what to do in any hand of heads up limit holdem. Doesn't matter if you're playing $1500/$3000 stakes, as long as you have 10 second to make an action, you can play UNBEATABLE* at heads up limit holdem. Right now. On online poker. Any site.

How soon before sites like stars and full tilt take Holdem out of their mixed games rotation? Hell, how long before limit Holdem is taken off the board entirely?



*unbeatable = not necessarily the best strategy against all opponents, you could win more by exploiting obvious tendencies, and also if someone else is using the same strategy as you, you will both lose bad to the rake. but you will not get beat.

lol wow it has to be a troll. no "poker writer" would be so poker illiterate to actually write that and seriously believe it. the guardian got trolled imo. the bad thing is any non-poker players reading that and believing it.

Thank you. I only brought this up because the casino heads up machine has been known to make this mistake. I assume this occurs because even with trillions of hands dealt, very specific situations may come up only a few thousand times and in those few thousand the computer did very slightly better by folding even though its EV was very slightly better if it called. Or something along those lines. (Am I right?) But in the case of drawing to the dead nuts with one to come it seems that a simple instruction (though maybe not that simple because of the unknown two cards) could guarantee avoiding an embarrassing (as far as the press is concerned) error.

How long did it take to get this through the review process at Science? I imagine there aren't too many people working on computational poker, so I'd be interested to know if any of the reviewers had any major hang-ups or demands for a revision.

Pretty amazing accomplishment, congrats! That Guardian article had be dying over here lol

Awesome work!

It does seem like only a matter of time until theres 3 4TB torrent files with the tables floating on the webs, and some software that lets you retrieve the numbers.

Set up some 6TB harddrives, adjust an existing bot, and then you've got Cepheus printing money (or Stars printing money for Cepheus vs Cepheus)

over/under 1 month?

No money in HU LHE, computer's solid

its been pointed out many times. there isn't much money to be printed at online hulhe. certainly not worth the effort of going through all that to make a functional bot that can play on poker sites. and anyone stupid enough to play against it long enough is also probably stupid enough that one could easily beat them without the bot.

fyp.

Leading quantum physics expert Sean Carroll wrote an article about this today. My bad if it's already been referenced itt http://www.preposterousuniverse.com/...tially-solved/

Whats next for the Cepheus team?

Really interested in this.

And also, this.

Luls

If I was in the mood the gently rib Sklansky, I would say that computer scientists prefer to solve toy games like Kuhn poker and Limit Holdem because real world games like NLHE are too difficult to solve with present techniques.

A technical quibble here is that you can't just copy one of Cepheus' actions and consider it unexploitable. For the strategy to work EVERY decision you make has to be a duplicate of Cepheus', which includes taking actions with the correct randomly generated frequency when they aren't binary.

If I'm reading the CFR+ description correctly, it evaluates the entire game tree on every iteration, so it should learn to never fold the nuts on the end very quickly (after first iteration?). Folding (occasionally raising) hopeless hands on the end shouldn't take much longer. Once it learns a reasonable river strategy, the turn decision with the nut draw will follow since the algorithm evaluates the EV of every possible river card on every iteration.

So if I'm understanding it correctly, CFR+ is less susceptible to this type of error as Monte Carlo-based algorithms which may not sample some nodes frequently enough. Of course, if the decision is very close after accounting for implied odds, it might still get it wrong. But then it wouldn't be an obvious mistake.

nburch - Can you confirm? Also, congrats!

The Guardian is the new Daily Mail.

Also congrats!

If you mean that capping at all is noise:
Can't you solve the (ever so slightly) easier game where BTN is not allowed to cap, and when finding that the BTN's average gain is no lesser than the original gain, conclude that capping was useless?

@ChrisKPHall: "Sadly I only got literally two hours to write, play and turn in the article and some parts got cut."

[x] Counterfactual regret minimization.

I'd like to see it try to play against a real quality human player. My money is on the computer. No tell.