Quote:
Is it that its not a mistake since by definition rounding will choose the high frequency play?
As long as your opponent plays GTO and doesn't adjust it's not a mistake to choose any >0% frequency play all the time. If there is mixing in GTO all the plays have the same EV so it doesn't matter which one you choose. It only matters when your opponent starts adjusting.
So if we, say round a strategy and the computed frequency goes from say .21257 to .25 we're technically slightly deviating from the GTO strategy. Therefore we're more exploitable.
Quote:
It's not necessarily a mistake to enter a low-frequency node. In fact, even if a strategy option has a "low" frequency (and I think the definition of "low" is arbitrary) we still need to choose that option with that corresponding low frequency to play a perfectly GTO strategy. So if the frequency is say 10%, we can't just totally ignore that option entirely--1/10 times we need to choose that node in order to be play perfectly GTO. (Again someone please correct me if I'm wrong--still learning the ropes)
It is a mistake, however, to enter a zero frequency node. Those nodes are unilaterally lower EV compared to the other nodes in the tree.
Yes, the key here is that while incorrect (in comparison to GTO) mixing makes you exploitable you are not losing EV as long as your opponent doesn't adjust so rounding itself (without adjust from the opponent) doesn't lose any EV.
Quote:
How are preflop hand orders determined?
The ones built-in are taken from some solutions I've run long time ago. You can create your own though.
Quote:
What's a good bb/100 to solve for for preflop trees? Is there even a good answer this this question?
2-3bb/100 is good enough for most purposes although the current version of the solve is able to get much lower on most trees.
Quote:
Why does the preflop solver give exploitability in terms of bb/100 and not a percentage of the pot? Can you easily convert between the two metrics?
It always gives exploitability in chips per hand. Those you can convert to either bb/100 or % of the pot by using simple arithmetic. For preflop it just doesn't make much sense to use % of the pot as it's not clear what starting pot even is.
Quote:
In other words, given an n-flop subset can we determine how well those n flops approximates the overall game?
Yes but to know that you need have solutions on 1755 flops to compare total EVs or EQs. We do have some and we run some tests on it. Some of those results were published in the old blog post about preflop orders here:
https://www.piosolver.com/blogs/news...the-whole-game
Quote:
Any suggestions on how many flops we need to fit into a preflop tree to get a "good" approximation?
That's a judgement call. Many very smart people have different opinions on it. You get something sensible starting from 40-50 flops and maybe it doesn't make sense to go higher than around 120.
Quote:
What exactly do the weights in the subsets indicate?
They are probability. The bigger the weight the more this flop counts for overall results.