Re(5): Ubilla 1st game singles... Posted on January 31, 2019 at 04:45:41 PM by Craig G
Just to be accurate, he didn't own the "Jai-Alai Supercomputer" - only had temporary use of it.
Wow, that is hilarious. I thought that the 'supercomputer' part was only used as an exaggeration.
All these years of suffering with, "Mirror, mirror, on the wall... who has the slickest jai-alai analysis software of them all?" always led to, "Well, in the exact calculation department, it's Lorenzo!"
But he was doing it on a supercomputer?!?
Cool thing is tho, that thinking he was doing it on a desktop computer kept inspiring me to match or surpass it, and I finally did. I use my genome to churn thru every possible game and get the exact probs, GIGO, be damned.
Extra precision isn't needed at that point and, being a sim, it'll never be exact, anyway.
So what if they're off by a percent here and there.
I am really big on proving / testing everything. Empiricism++.
If I use my best sim at 1 million cycles and examine the top 20 tri list, I usually see some jiggling. In fact, it's amusing to look at my 'Kruncher' which delivers exact output, look at the tri ranked #21, and predict that it will jiggle in and out of my sim top 20 list. Yeppers, it works.
So, if "Jiggling has been ruled" for a top n list using 1 million cycles, what do you think will happen at your 32,767 level? BTW, even at 1 million a 10% freq variance for some top tri's is not uncommon. Like I said, I'm, in a position to perform the tests.
Regardless of that, I agree that the practical limit is more in the area of our ratings accuracy.
But I still like the idea that for a given betting package, eg top 14, the selections will not vary from run to run.