Poker-AI.org

I'm new to this optimal stuff, and 6 max isn't normally where people start... but hey.

My algorithm develops an unconventional UTG strategy, which is limping a fair amount - including some big hands, usually KK / AK. By introducing the big hands into its limp range, it means it can widen it's limp - 3 bet range, which in turn means it can limp a lot of marginal hands with less fear of being re-raised. Interesting thought process I thought. It currently plays ~20/10 UTG, which is far more loose passive that what's normally recommended. Maybe this has some mileage in human play?

Having said all that, it's also not beating my exploitive bot yet... I'm wondering if that aspect of the strategy will remain as it converges.

Well..I can certainly understand the limp reraise with big hands. I think it would be ok to do it once in awhile to mix it up and certainly if there is a big LAG dude behind you or on the button. I think in general its a bad idea and you should be raising all your hands first in from any position. The last thing you want to do is be out of position in a multiway pot.

Without knowing the algorithm/model used, how can we be sure this isn't due to accuracy error?

We can't... I still need to do a lot more work and a lot more training. I just thought it was an interesting strategy when I noticed the reasons behind it.

I never believe that openlimping big hands like AA is more +EV than openraising them.
You can coldcall them for sure once in a while but never openlimping.

Be aware that both calls have much more variance and you will be multiway postflop more often. -> your bot has to know when to fold the Overpair now

I agree, but the lost EV could be small, and more than made up for by the +EV from the increased range of marginal hands it can play from UTG. The bots limp 3bet range is pretty big, and that can only be achieved by limping the top hands occasionally. It makes it far less predictable UTG.