Poker-AI.org

Hey guys,

Does this look like something that could be turned into break-even / profitable?

It is playing on 0.01/0.02 NL Zoom @ Pokerstars. I reached Silver star in less than 30 days and still not flagged.
Roughly lost 90$ in 115k hands with a rules based bot.

I can give you more stats from PT3...



Jose.

I do not see why not. It looks like you have to just work on the AI a bit more. I have a Zoom bot I created last year that was generating 8BB an hour at the same stakes. It dropped down about 3 BB at .25/50. I didnt try it at higher levels. I didnt think it would have won at anything higher and my thoughts on game theory radically changed so I dropped it and moved onto HU.

What I decided that it was better to play at low stakes and try to increase the hands per hour to make it worthwhile. You should be able to get 275 hands per hour average per table if you utilize the fast fold button correctly. So you should be able to play 4 tables at 1000 hands per hour roughly total which is their limit (it may have changed by now..the more the better obviously). My goal was $1 per 60 hands played and I surpassed it at $1.50 per 60. My AI was totally small ball which I disagree with now but for micro stakes it worked very well. The great thing about Zoom is you can make the bot very predictable in spots without any consequences.

So just work on the AI more. Mine is also rules based and it took about 2-3 months of tweaking once I had a decent general strategy coded (a few weeks). One thing to watch out for in Zoom (and others) is the all in situations..make absolutely sure you have side pots correct or your EV will be way way off. I ran into several botters that had forgotten about that oddly enough. Its not easy to figure out when you have multiway situations.

Show bb/hand won. If it's a static strategy you can probably expect it to get worse over time, even at Zoom.

Your bot lose a lot at hands without showdown. You should find some spots to bluff at. Does your AI CBet enough for example? Of course bluff spots are tough to play and you have to take care to not make your bot plays worst on those.
Your bot might fold to much vs bluff too.

More info please.

The idea is simple - you should find the typical situations where you bots loses money. And it is not about vpip/pfr/any_other_stat thing.

For example, it loses a lot without showdown. Find the street where is happens. Probably it is preflop. Find, if it is blinds or bot fold to 3bets. If blinds - defend them or steal more and so on.

Also you can try to raise the blue line and win more on showdown, but I think it is more difficult.

_________________
Бетономешалка! Мешает бетон!

I noticed for Zoom that its important to play very loose preflop. You should be seeing the flop about 25% of the time minimum and raising first in from any position with a wide range of hands. I am not sure if your playing 9 or 6 max but you should have two strategies for shorthanded and multiway post flop scenarios which are vital. You should also be avoiding any near coin flip situations on the all-ins to reduce variance. Some people may have issues with that but I have found it to be better on the bankroll.

From your graph it looks like your way too tight as everybody has mentioned. Marginal holdings and bluff steal attempts are where the money is at. The problem with that is they only work shorthanded so your gonna have to do more raising preflop to isolate and also call more "bustem" type hands multiway so you get into more situations to flop a disguised hand in a big pot. I would rather have 75s then AA in a big pot.

How much rake is paid?


probably $50/m maximum right now

hello, sorry for posting to this old thread, but i have one question to shalako. I am also targeting zoom 6max, and am still not sure whether to go the simulation based or the rule based way. Till now i have spend a lot of time with simulation related stuff, and the more I dig into this the more i see all the weaknesses and am considering if i should just switch right away to a rule based approach(i dont have a finished AI yet). So here my simple question just to get an idea what is possible with rule based approachs:
Shalako, you wrote you make 3bb an hour at .25/50. Do you mean that is 3bb per 1000 hands, as you write you play 1000 hands per hour(which would be very marginally winning)? because below you write you make 1.50$ every 60 hands(which would be absolutely awesome:)). Sorry for digging into this, but a clearification would help me a great deal with my decision.

Hey. It won 3bb per 100 hands played or around 9bb an hour as Zoom is like 300 hands per hour if your utilizing every possible use of the fast fold button. This was like 3 years ago when Zoom first came out. I doubt it would do well now but maybe.

I would def go with a rules based AI. You have two options with rules. Either a hand type rules approach like "I have top pair and a draw" or a GTO approach using only equities. I have heavily adopted the GTO approach lately for PLO which I think is better. A mix of both is actually ideal because straight equity does not always work for the correct play. An example is a Qhi Flush. Raw equity wise its great but you really cant feel that great about it when your faced with a raise. This is where bots have problems. No matter how you look at its +EV but with action its a fold. So this is where a mix of hand rules and equities is needed although there are ways around this. One method is too run equity not against an entire range but against the nuts (or a range of > X% equity) to figure out your worst case scenario. So if you still have +EV vs the worst case scenario with pot odds your good to go. This works ok for marginal made hand situations but if its overdone you might actually be losing EV. None of this is easy to balance and your going to make bad folds on occasion but that is poker...

hey, thank you very much for the elaborate answer! Mixing rules and some equity calculation and eventually even simulation was what i am thinking of. Again, thx a lot!