from Amazon
Tuesday, December 09, 2008
UZR v PMR v UZR
Dan The Turk compares PMR to bUZR (both use the same data source, but obviously different engines). Dan lists the top 10 in differentials.
Soooo… I took those guys, and compared them to sUZR (UZR using STATS).
Iwamura: almost perfect match to bUZR
Feliz: 3 runs away from bUZR
Glaus: perfect match to bUZR
Renteria: much closer to PMR
Reyes: perfect match to bUZR
Hawpe: almost perfect match to bUZR
Mora: almost perfect match to bUZR
Torii: half-way between bUZR and PMR
Uggla: closer to bUZR, and even further away from PMR (all other cases, sUZR was between bUZR and PMR)
Abreu: perfect match to bUZR
First of all, it sure doesn’t look like bUZR and sUZR are that different based on this non-random sample. Which reminds me that “correlation” is not what we want to test, just the average difference or RMSE. Correlation will alter the slope.
In any case, it certainly looks like MGL’s engine behaves very similarly. It looks like there’s only a couple of runs difference either way. MGL: can you re-run your bUZR/sUZR comparison, but look at the average difference, and not correlation?
And clearly, Pinto’s engine is doing something very different from MGL’s.