Quote:
Where do you get there being 1833 unique flops? That isn't correct, fwiw.
Wrote my own python script.
Order hands by card value.
Then fix first card suite to heart.
If in original flop first card suite was equal to second card suite then set second hand suite to heart else diamond.
something similar to 3rd, set it to heart,diamond or cross depending on original flop.
And so I get 1833 flops. I know theory says 1755, did a compare and I have 78 flops like XhXdYd that are perfectly equivalent to XhXdYh. So there is the difference (9h9d4d could be encoded to 9h9d4h). In my code I have the right frequencies for those flops so there should be no problem there
Quote:
You say you're using bucketing, and then you say that your results are better because you compared your own sims ran on smaller subsets to full subsets. I'm way more suspect of your bucketing techniques (I've only heard of people doing this wrong) than small flop subsets.
Indeed two problems with bucketing, one is doing it right and the other is using the buckets in the game tree as hands in the same bucket on flop might land on different buckets on turn so the game tree becomes a digraph. So far I checked out my clustering and I think I got an ok-ish bucketing. As for the second part this is what I am mostly proud of, I devised an algorithm that goes in 4 passes (one for each street) that so far seems to work. I understand the suspicions, it is something that is hard to get right.
Quote:
FWIW there have been plenty of comparisons of small subsets (~80) to big subsets like half or even all flops. Obviously there will be some bias in the smaller subsets, but they compare surprisingly well. So, seriously, if I had to put my money on anything it would be bad bucketing.
What if the bucketing is done right ?
Would not this get better results with all flops sets than using a subset and no bucketing ? Just asking!
Quote:
Oh, and the best-response calculating. There are 3 major competing programs in this field and they are all in agreement about their MES calculating. I'd say it's safe to trust it's bug free.
What is MES ?
Wasn't talking about those programs, as a programmer I was saying that if a solver uses the same code for solving it and computing best response (with small changes) than a bug in the code would lead to buggy best response.
On a side note, I sense a lot of negativity here and can't explain. I am not here to badmouth piosolver or GTORB, I respect those programs and their creators because as a programmer I have a good idea what goes on in working on such a beast. What I have is a research project of mine that I worked for past two years fulltime and a bit more before that part time and for the first time I found an algorithm that allows computing an equilibrium using all flops, not a subset. At the moment it reached a very stable phase and can be developed in further directions and I love working on it. You guys should be happy that in addition to existing solvers there is another project that tries something different. Lighten up! If anyone would be willing to get involved/be a backer for this project then it's good, if not it's also good! I appreciate the talk about GTO in this thread and I love responding, but in some posts I feel like I'm being put to wall in front of the execution brigade! Maybe it's my fault for phrasing some things not in the right way, but hopefully I explained things better now!