Ive never gotten to the second hump on any camel
"Your honor, with all due respect: if you're going to try my case for me, I wish you wouldn't lose it."
Fictitious Boston Attorney Frank Galvin (Paul Newman - January 26, 1925 - September 26, 2008) in The Verdict, 1982, lambasting Trial Judge Hoyle (Milo Donal O'Shea - June 2, 1926 - April 2, 2013) - http://imdb.com/title/tt0084855/
Yes, the system is complex, but let's not over-glorify what is happening and why it's taking so long.
It's not the sims themselves which are taking so long, it's the development of the software. This is mainly due to the diligence we are taking to ensure that the results are credible and that Tarzan is completely confident in their accuracy. I summarized this earlier in another thread, but here it is again:
- I first reproduced some Hi-Opt II results from CVData+CVCX in order to demonstrate that my software framework is correct. Norm and Don were very helpful during this phase as was Dog Hand. The other reason we used Hi-Opt II was because it is the system we chose for initial comparison.
- I then implemented Tarzan's system in 3 levels: T Basic, T Expert, T Expert with Key Cards (need a catchier name for that last one!).
- At each level, Tarzan has reviewed logs for the playing of each hand (thousands of lines for each) to make sure that the internal logic is correct as well as the final decisions.
- After some initial results of the final level, Tarzan realized that some calibration of the key card adjustments were in order. That's where we are now, although we believe we can see light at the end of that tunnel (or is it a train?).
Stages 3 and 4 are what have taken 99% of the time.
While this is true, at the end of the day, each simulation uses the standard 20 billion iterations from BJA3 Chapter 10. My software is not nearly as fast as Norm's. Each sim takes a couple of days, where it would be a matter of hours for CVData+CVCX. Still, lack of computing resources is not the reason for the length of the project.
Nope --- it's just me
Once again, most of that time was spent enhancing my software in order to handle the complete range of games of interest. My version of the "beat it to death" algorithm for index generation is not as fast as Norm's. It took 100 billion iterations for those hundreds of indices to converge which took about a month (per game) on my spare laptop.
On top of this, I am doing this in my spare time (I have a full time career and a family) and also doing work for others.
Once the results are final, it will be up to Tarzan what he publishes and when. I'm not trying to make excuses. I just want to set expectations.
Well, the actual performance of the system has been the subject of many speculative claims over the years, but it has never been simulated in order to produce a SCORE, until now. The final conclusion will boil down to the usual balance between benefit and complexity.
... and whether your bankroll is large enough
Bookmarks