Heuristics and maths

I found an interesting article that I think is worth more people reading and discussing

https://towardsdatascience.com/poker-is-all-about-heuristics-not-math-8603e69527f2

What I got from it, in a very broad sense, is that the “deep” maths of poker is very interesting but, basically, useless in a game situation. The author, I think, says that nobody can play as perfectly as a solver or some other tool, so we need to reduce all the fancy maths down to some or a few generalisations that we can quickly and easily use while actually playing.

I obviously have far too much “study” time since I found, soon after reading that article, a website by someone who seems to take data analysis far too far! A lot of it is too deep for me but there’s an interesting video that seems to be highlighting, unintentionally, the points of the article referenced above.

In the video, Lukich analyses a particular hand (or two?) using solver software. What I found interesting is that Lukich isn’t looking for ways to play closer to GTO so much as identifying places where a player seems to be deviating, with some regularity, from the GTO solution and suggesting that an opponent could use this knowledge to exploit that player by also playing differently to the GTO solution.

Maybe this is old news to more experienced players but it’s very interesting to me :slight_smile:

I hope this is interesting or useful to other people as well.

Regards,
TA

4 Likes

I just noticed this post, and I had already read this article (I work in the data field). Human cognition does have some key strengths (grouping similar items together to make connections) and weaknesses (computing the exact EV of every possible action with every possible hand in the ranges of three players on the turn). Without extreme real-time assistance humans cannot make correct GTO decisions the way solvers do. We solve the problem by bundling solver solutions in ways that we can actually use, like hand types, board types, positions, stack and pot depths rather than precisely knowing every variable.

That is why trying to adopt gto strategies (and adopting any poker strategy really) can lead to losses and mistakes. It’s just tough to know if we are actually playing how we want to play and whether our decisions align to our strategy as planned. That is the value of off table study and solvers. It is interesting stuff and hopefully will prevent nlhe from being a completely solved game in our lifetimes…

2 Likes

Thanks for posting these links, TA.

The key lesson in the article is that, human versus human, observation of individual and group tendencies is the most practical (even if sub-optimal) approach for playing consistently winning poker. “Exploiting the meta,” as the author describes it, is particularly useful on Replay, where loose-passive play (e.g., preflop limping) is the norm.

I prefer learning the game from accumulating experience (i.e., playing) rather than doing off-table work with solvers (i.e., experimenting), which is unnecessary (though certainly instructive) if you are not playing for actual cash, which I don’t do. Besides using odds calculators to review hands where I may have made borderline -EV bet/calls, I find it more instructive to use a heuristic approach to breakdown mistakes in hands I lost. Usually, it’s my own mistakes that cost me chips, and these mistakes tend to be observational failures or other weaknesses unrelated to math.

The type of real-time hand analysis presented in the video is definitely useful to incorporate even into recreational poker, though this type of analysis assumes that opposing players are always making thoughtful decisions, which is not always the case in free poker.

1 Like