If you're trying to lecture me on solvers can you please use the correct term?
The algorithm in solvers is called CFR and that's not the same as EV regret defined by the software.

Here's a quote from the creator: FE himself defines EV regret just the way i said.
https://snipboard.io/Nisx3b.jpg

(and yes that method of simplification is awful. sorry)

Okay I think we are getting somewhere. When does the bucketing happen? Before or after the sim is ran?

edit:
to make it crystal clear: I am trying to say it matters whether the bucketing is done by the solver or the human.

CFR - counterfactual regret minimization, minimizes EV regret.

What FE is doing is assuming that the solver will continue to play the original fixed GTO strategy (passive exploitation), instead of specifically trying to exploit the imbalance (active exploitation).

https://jonathanlittlepoker.com/exploitativevsgto/

But even that doesn't work. Look imagine you try to simplify on the flop by bucketing all your top pairs into betting pot.
Now we go to the turn.

Does it or does it not matter for our turn strategy whether we have 25% or 100% of our top pairs?


I'll skip the J little article. I love the guy but even he'll admit that i could teach him a thing or two about poker.

That article is actually taken from Modern Poker Theory by Acevedo, JL is just paraphrasing it for his readers.

Look I take your point. If someone naively uses this in a vacuum with a high threshold they will miss the fact that they may become quite imbalanced on later streets. The tool is meant as a simple metric to measure mistakes of bucketing simplifications (vs a fixed GTO strategy), designed to be used in conjunction with other metrics. It's still a useful metric imo.

If you want something that punishes mistakes, you could play "range vs range" against a MES solver that will adjust to exploit every imbalance. I tried this once on RangevsRange.com and it's excruciating. I actually think such a tool could be extremely useful, though most players aren't accustomed to constructing entire range strategies each move.

It's not just a balance issue though. It was a genuine question. Do we proceed as if we had 25% of top pairs or 100%? I know the solve in front me assumes "gto" play. Yet in reality i have a different range. Sounds like a bit more work might be in order here...

In order to do what FE proposes properly you have to:
nodelock your opponents optimal response vs your simplified strategy. Then calculate the ev of the new game. This is the only way to get your true EV loss. This however requires an entirely new solve. And you can't shortcut this.

And you can't resolve from within GTOX either...

Fwiw this is a great exercise to do if you want a simplified preflop strategy but I am getting off topic here.

EV regret as defined by FE is nonsense plain and simple. If it was easy as grouping your hands together after the solve I hope i'd have thought of this many years ago


As for the range training idea. I love this. Wanna DM and tell me how this works exactly? I have a few questions and perhaps can get a similar tool build fairly quickly.

For sure I'll shoot you a message. I actually really want to see someone make a tool like that.

As far as bucketing goes, there are essentially two ways to measure the loss of "mistakes":

1) Assume villain will play fixed GTO, and will not adjust - Mixing mistakes will not get punished, only pure mistakes
2) Assume villain will adjust to exploit every imbalance - all mistakes get punished, but the user needs to construct an entire range strategy.

1)But i still don't even know whether i am supposed to play the turn as if I had top pair 25% of the time or 100% of the time. Without that information I can not determine my GTO strategy for the turn.
Or is the implication here that this doesn't matter and we should follow the old solve that no longer represents our strategy?
Because in my opinion you can't do that.

2. well that's obviously worse for us and I'm already arguing 1 is bad

1) Is actually built into the calculations of a solver. That's how you can see the EV of actions that aren't taken.

It's relatively trivial for a solver to calculate the best strategy against a fixed opponent strategy, even if your hand never reached that node in the tree. Old school Pio used to be able to show strategies for hands/lines that are never taken.

Yes it's trivial but you still need to do it. GTOX doesn't give you the option to do it.

Would you please answer my question: I follow GTOX. I bucket my top pairs into betting 100% only on the flop.
I now wanna look at my turn strategy after i bet pot on the flop but their turn solve assumes I follow different frequencies. Is the answer GTOX gives me still correct? Yes or No?

Yes, assuming the villain's strategy is fixed. Most training software allows you to play out lines that aren't taken.

"The simplest explanation I can come up with is that EV of a hand is not affected by the composition of our range, it's only affected by villains strategy. When we say our hands EV changes when our range changes, what we are really saying is that villain will adapt to our new range and therefor his strategy will change, and the EV of our hands will change in consequence."
-Aner0

https://forumserver.twoplustwo.com/1...23/?highlight=

So suddenly having 4x as many top pairs in my range does not change my strategy nor EV? And it doesn't change the ev of all other nodes either? What does it change then? And in which line do i allocate my extra hands on the turn?
In other words you are saying we can pretty much do whatever we want if villain responds gto?


wow it's all so much easier now. I'll group all my hands together like this. GO FE! thank you for opening my eyes

while this post was clearly sarcasm so far this part isn't:

PIO removing unused lines is a fantastic feature. And GTO trainers should not allow you to study unplayed lines.

Against a fixed strategy, You simply optimize each hand in a vacuum against villain's fixed strategy. Changes to your range have absolutely no effect on their strategy. The EV of each hand can be calculated in a vacuum. It really is that simple.

You can completely disregard your range in this scenario, as all that matters is villain's actual fixed strategy.

If you don't understand why, then you may want to read that thread above, slowly and carefully.

Okay I got it. Their strategy won't change. This has been clear from the start. I agree that the EV of a single hand will the stay the same obviously. Top pair will win the same amount of money.

But what about our strategy? Does it not change either?
Where do the extra 75% of top pairs we now have go?

Doesn't our EV fall of the cliff in other nodes of the game tree since we no longer have top pair in those nodes? (while of course improving in the bet 100% node since it's where all the top pairs are)
How is that effect measured?

It actually doesn't matter that you cap yourself on other streets since the fixed strategy won't adjust to exploit it. These hands can continue to play exactly as they were before without losing any EV.

Hands that would have got there at any frequency don't change their strategy at all. Because again, you're playing against a fixed strategy.

Hands that would have never got there (0% freq)... well that's a bit more complicated as it depends on the exact implementation of the solver. But the EV and correct moves can still be calculated for these hands in a vacuum against the fixed solution.

If you've ever used a solver that showed you the EV for an action not taken, it has calculated exactly this. Though it may not show the strategy for 0% moves for aesthetic reasons.

I am NOT talking about the ev of a single hand. I am talking about what happens to the EV of our checking range If we never have top pair in it?
Will it drop or remain the same? Of course assuming villain plays a fixed strategy.

The EV of whatever hands remain in that capped range won't change at all.

Of course if you remove value hands the total range EV will go down (that's pretty obvious), but you won't lose any extra EV beyond that. Every one of those crappy 2nd pairs and K highs will realize exactly the same amount of value as they did before.

Now if you changed villains' strategy to something that exploited this imbalance, your capped range would get absolutely destroyed. They wouldn't realize ****. That's the point I'm trying to make.

Right so in summary against this kind of simplification we have:
- it messes up frequencies
- it hurts our EV in other parts of the game tree
-we can no longer look at any turn solve as the turn solve does not account for the simplification we took
-and top it all of: our simplification is painfully obvious so getting exploited is a real thing.

NGL this sounds awful and far worse than restricted betsizings. Atleast with restricted betting options one can accurately measure the EV loss of the strategy they implement.

• it messes up frequencies - not against a fixed strategy.
  
• it hurts our EV in other parts of the game tree - Against a fixed strat You just moved those high value hands from one area to another, and the remaining capped range doesn't lose anything extra.
  
• we can no longer look at any turn solve as the turn solve does not account for the simplification we took - depends on your solver. If the hand takes the line at any frequency then the resulting strategy on following streets is still correct (against a fixed strategy).
  
• and top it all of: our simplification is painfully obvious so getting exploited is a real thing. Depends on the extent of the simplification, and if they can correctly adjust to exploit you.

I don't think we're gonna agree on any of this, but thanks for the spirited debate regardless. Best of luck.

Maybe you guys can shed some light on this.

I ran two sims for a 3BP, SBvBTN

Sim 1 (GTO) - OOP bets 25,33,40,66,75,All-In
Sim 2 (Simplified) - OOP bets 33,75,All-In

In both Sims, each player has 2 raise sizes (50 and 75)

Turn and River sizes for OOP are 25,66,100,All-In

IP has the following bet sizes when OOP checks

Flop - 25,33,50,All-In
Turn and River - 25,66,100,All-In

In the screenshot below, the top 2 images are OOP's flop frequencies, and the bottom 2 images are IP's response to a 1/3 c-bet.

The frequencies don't really change all that much. So what is the benefit of solving a hand with so many sizes on the flop?

@Newguyhere
The frequencies and the strategy will likely change more later in the hand, although both sims look reasonably complex already. Care to share a few river examples between the two sims?

Here's another try. Simplify the flop to the most frequent size, and see how that changes the strategy down the hand.

I actually do think we can agree on this. All it would take is some nodelocked sims that show, that we'll get the same results with that method as we do with FE method correct?

I even agree with the first point right away. Weakened ranges will lose some EV. Strengthened ranges will gain some EV. But what matters is the overall EV of our entire strategy.
Whether that changes is up for debate and can only be settled with a pio sim

As for the last point. I gave you my example: pot betting with 100% of top pairs.

@newguyhere
simply put there is none. You can pretty much feel free to choose your own set sizings. Just make sure you can actually implement it. (so use just 1 or 2)

If you do use 2 on the flop make sure your bets are more spread out.

overall feedback for your sim. You gotta cut those sizings down. A lot... and space them out a lot more. 50% can't be your 2nd largest bet

You'd have to nodelock every action on every street to truly test it but yes.

Let me make a sweeping generalization:

If a hand mixes between two or more actions, those actions will ALWAYS have the same EV regardless of how you mix (against a fixed strategy). Anything else is just solver noise from not solving down far enough. As far as I'm concerned that's a hard law of equilibrium.

The only time mixing mistakes are punished is when your opponent adjusts to exploit you.

Do you agree?

Here you go. For the turn, I just picked the size used most frequently. And not sure if it matters, but the flop is Qs9d5d




I will try this next.

This is 100% correct

And here is the GTO vs Simplified comparison when simplifying down to the most frequently used size on the flop.