Theres been some recent talk of this algorithm scattered around the STT strategy forum, but I though it made sense to consolidate some of my thoughts/findings about it here...
Having read Ben's paper and his recent article here:
http://www.pokericmcalculator.com/en...rts-icm-model/
I have to echo much of what's been said here... The algorithm seems very interesting and clever, but many of assumptions and conclusions are fairly eyebrow-raising...
All the claims of accuracy and "truth" of the equity results are only based on a different, mostly arbitrary, algorithm that also lacks any kind of observable connection to reality. I know this is technically stated in the article, but then saying about the sample hands: "Therefore, these simulations suggest (according to my intuition anyway) that the BR algorithm is not only more accurate with respect to the ICM model but also more reflective of real equities." is pretty far out there. If your intuition can tell you that a calling range of 78.9% is "more reflective of the real equities" than a 69.5% calling range, then your intuition is much much better than mine....
Also, when you say:
Quote:
This ignorance of realistic factors is a common criticism of the ICM, however (as far as I know) no viable alternative model has been implemented that does account for such complications.
There is my "Predictive Simulation" model that powers SnG Solver... It accounts for blinds/antes/position/future rounds ( full disclosure: SnG Solver is a commercial app, I wrote it, and therefore everything I say is biased
)
Nevertheless, this algorithm is intriguing and I've implemented it and run it through my Nash STT simulator. I set it to play 6-handed (3 players playing a Roberts-ICM based Nash push/fold strategy, and 3 playing a Malmuth-Harville-CM based Nash push/fold strategy). After 200k simulated games, I got this:
So overall, this shows the Robert-ICM with -0.36% RoI vs the MH-ICM. So, at least as far as short-stacked, super-turbo style STTs, I cant say that the Roberts ICM is an improvement.
EDIT: BIG CAVEAT: plexiq is also running a sim like what I've done and has told me in PMs that his results differ quite a bit. In the past when we've compared sim data, we've been very very close so one of us may have some kind of systemic error. I will be auditing my results and will update if I find anything. I'm sure we'll see some more posts on this.