What is the point of 5 ****ing sizings, someone please fill me in? It must eek out like 0.001bb/100 compared to 2-3 sizings while making the sim damn near impossible to study. i dont get it

I do this not so that I can know how to balance 5 sizings (nobody can really) but to find what sizings the solver likes to use most.

After that, I rerun the sim using those popular sizings and I should have 1, maybe 2 sizings in the finished result which is *ideally* losing minimal EV from the optimal solution.

Feel free to tear me a new one if I'm wrong about this methodology

Nah that approach makes perfect sense imo. That's the human and practical/actually applicable approach.

That’s how I use PIO as well, but that’s not what these products do.

They’re selling closed solutions for different spots. You can’t modify the bet-sizing options.

The product is probably crap for practical applications because you can just use 2 sizes instead of 5 sizes and suffer minimal EV loss in many spots.

Are there examples of poker pros who use 3 or more flop sizes? Maybe Linus or Foxen use 2 flop sizings in some spots. But is there any evidence that pros are using 3 or more sizes?

The fact this app uses 5 sizes does tend to make me think they’re targeting the app to be used for RTAs, ie cheating, though they obviously deny it. Rather suspicious now that I think about it.

Exploitative-minded poker players often use several flop sizes

I was asking about GTO-based tournament pros. We are discussing GTO sims after all.

Dom needs to realize that publically bashing his competition like this is poor business etiquette.

• DTO's approach to solvers involves high accuracy and fewer bet sizes. In other words, they want a really precise solution for a less accurate representation of the game space.
  
• Odin's approach involves tons of bet sizes with lower accuracy. In other words, they trade a less precise solution for a more accurate representation of the game space.

The tradeoff between these two approaches looks something like this:




Odin claims to solve down to 0.5% dEV. That means it can be exploited for, at most, half a percent of the pot. The thing people don't realize is that there are nearly infinite different strategies that are all extraordinarily close to unexploitable. If you want consistent strategic outputs across different solvers then you really do need to solve further down, but all that tells you is that there are many playable strategies at that node.

True GTO probably involves infinitely many sizes. So why stop at 5?

Anyway, if Linus cannot balance 3 flop sizings what is the purpose of looking at a sim with 5 flop sizings? Seems bad.

It’s like the zenith preflop ranges. They involve at least 3 open sizes and 3 raises sizes for every configuration.

Nobody can balance mixed strategies with 3 preflop sizes.

Nobody can even balance 3 flop sizes with mixed strategies.

What’s the point of complication if it cannot be memorized in the human brain.

That is, unless the point is to use the sims for an RTA.

What's the point of trying to memorize solutions?

Even with one sizing at each node there are literally millions of different situations (boards/lines/positions) that could come up in a standard cash game. You can't memorize all that. The point is to study GTO principles not memorize the solution.

Personally, I would rather have access to something with several sizes so that I know how to respond against different strategies and could research trends more accurately across the game tree.

Any link? Would love to see this

I didn’t mean to imply that anybody would memorize the solution directly. Rather to study the sims preferred bet sizing and range construction on different flop textures, and then formulate a heuristic that allows you to construct an approximation of the GTO ranges on the fly in a live setting. Memorize the heuristics not the solutions themselves.


Okay, I see what you’re saying. Even though you might not use several sizings in your actual strategy, you cannot predict what sizings your opponent is using. So you want a sim that shows responses to all possible sizings. That makes sense, actually. Thanks for the explanation.

Dom and I were debating game theory the last time he was attacking his competition.

https://forumserver.twoplustwo.com/2...73/index4.html



I agree the goal is to build heuristics. I just find that easier to do with more data rather than simplified solutions. For example, looking at aggregate flop reports is a lot more insightful if your tree is complex. There are always tradeoffs, but both approaches are valid.

I've said my piece on the multiple sizing debate many times.
Look some people will disagree (hi tombos) and that's cool. I just prefer less sizings and more accuracy. In a way it's good for me that my competition prefers the other approach.

This below sums up my thoughts on that matter.
again i don't want to derail the thread but I find this utterly insulting.
1) i'd never run a pio solve with 5 sizings (fun fact they actually run it with 9 then round down) and put it in my training
2) here's the dtoish solve. (which i obviously didn't post cause it has less sizings and thus different results)
https://gyazo.com/2959716e4f620a9327f37dd19afc7dc9

what's wrong with the odin results? about 50% too much checking. Far too much mixing. Even the developers of simplepostflop are suspicious and think it looks like it wasn't solved properly.


I mean if you're still giving them your money after this I can't help you

could you post the turn/river sizes? it does look like someone messed up somewhere

This is Odin. The answer is all the turn and river sizings.


Fwiw i really don't think this issue is sizing related because when we tried to copy the tree for fun this is the output we got at 2.2% exploitability. And I do not claim to have hit their river sizings perfectly. Still: it looks a lott like the Odin solve
https://gyazo.com/3f020b891cf1716d2bacca897aa57288


that's hardly shocking either because 30bb BTN vs BB on KQ8 is going to be bet very often. Any Sim recommending this much checking is very suspicious

What are the chances that Odin set their exploitability parameter to 0.5 instead of 0.5% by accident?

I do think Dom is a bit aggressive in his approach, but publicly raising concerns does create greater transparency. Especially in a market where most people might be unaware that there even is such a thing as sim accuracy and its importance. If a lot of people were paying for say a chess learning tool and it turned out it was suggesting the 7th best move each time (not saying this is the case with Odin as I don't know) I feel that warrants a public discussion/scrutiny, because the average customer might be completely oblivious to it.

I'm personally involved with a competitor (Deepsolver, web-app solver, not a sim library) and while we do have the fastest solver in the world the trade-off is we have an average ND of 0.5% (meaning at most you can win 0.5% of the pot vs. its strategy). And to be honest I was a bit concerned whether that number would be too high for some more demanding people (it will be improved, but that's besides the point). As I understand the threshold among top players right now is 1% ND (anything above that is considered not worth studying).

Quite a few people are expressing RTA concerns in this topic - I believe Odin should get massive props in that department for self-imposing a 20second delay on their sims (I assume it was primarily done as an anti-RTA measure). It's an extremely difficult decision to make as a business. You do score some points with the community, but overall it costs you money (from regular, non-cheating customers who enjoy the speed) and puts you at a disadvantage vs. other competitors that don't have the same restriction.

Poker learning tools could and should be doing a lot more in that regard and collectively take a more explicit stance on RTA.

What do Deepsolver do to offset players using it for RTA?

Lol?

GTOWizard has built-in safety protocols if you try and use it as RTA and work with the sites that allow the use of the software off-tables. I can't speak for the other software tho.

about 50 percent

What safety protocols?

They have some sort of algorithm that looks at your timings and how you switch between sims to determine if you are using it as RTA. But there (understandly) isnt much detail about the specifics of how it works.

The creators are very upfront in their discord about it not being RTA and do work hard to combat it.

We've been vocal about RTA from the get-go. We've informed every major poker site about our app before our release so they could add us to their prohibited list. We've also communicated our willingness to cooperate with the poker community and the sites in case someone is suspected of cheating. Cross-referencing our sim databases with the date,time,decision/sizings in a hand history is pretty easy to do and serious proof in a case. Also we don't provide API so the only way to use Deepsolver is through our app.

As for measures that would prevent this from happening in the first place there is quite a bit we can and will do on our end, but I wouldn't want to get into specifics not to give any ideas on how to bypass them. I know it might seem like a cop-out answer, but unfortunately to a degree it's a matter of faith as in the case of poker sites having such measures as well.

That said after all input is given (stack/ranges/sizings) you still have to wait around 8-9 seconds to get flop results. That's tanking territory while 1tabling. However let me say this - if the other sim apps agree to do so as well we're more than willing to set a higher time delay (20 seconds or more). We will make this proposal public at some point, right now our social media presence is still growing. We might even self-impose a time limit independent of other apps provided this will be something the majority of our users will agree upon.

That’s a really interesting answer, thank you for responding. Hopefully every other app agrees to implement a 30 second minimum wait period.

Just to be sure I understood - your app notes the time that a player checked a certain flop, and poker sites can cross reference that with the time a hand was played and investigate the line that was taken? Very interesting idea if I’m reading it correctly.