- I ran some more test sets and tweaked more tonight. It's getting closer. I am
bringing things in slowly, but surely. By the weekend I should have a pretty
tight comparative data set that I can be happy with.
Just to keep you stat junkies satiated, here is a graphic with some test
results. The plot is NHL SOG/min on the x axis and test SOG/min on the y axis.
The cyan line is ideal, that is, it would be a perfect correlation if all the
points lay on the cyan line. This is the target I shoot for. The black line
and black equation are the least squares fit for the very first test set with
the draft disk. As you can see, with the 0.64 slope, there is a pretty strong
bias. Alot of the bias is focussed on the low end. Note how ALL the players
below 0.6 are above the ideal line. Also, there is some top end bias (low)
also. After a few sets of test/tweak, I've nearly eliminated the bias (0.915
slope, 1.00 is ideal). The remaining bias is focussed again on the very low
end, below 0.04. These will all be tweaked down in the next iteration. Also,
there are a few flyers on the top end, but these are hard to get in, as they
end up with SG hiddens over 100. Also, alot of these out of bounds top end
guys are players with very few games, so the variation from set to set is
huge. I'll try to tweak them in the best I can, but it's not really important
when they represent only 50 or so shots combined. This plot would be better
represented if weighted by SOG totals. But, this works for quick test/tweak
purposes. More news after I do more rounds tonight and tomorrow night.
- Just a quick update.....the disk is totally finished and in error
checking right now. The disk will be released as soon as we get a clean
bill of health from the error check team. We're close!!!