paper

[markhwhiteii]

• [x] write it
• [x] upload draft
• [x] cite draft, with link
• [x] submit article after package on CRAN, updating paper to include CRAN reference

[markhwhiteii]

tinker with the part of the algorithm that synthesizes the winning and losing scores. is standardization, flip the sign of worst, and averaging the best?

[markhwhiteii]

synthesizing methods tried:

• mean(std(B), std(W * -1))
• mean(B, (1 - W)) --> even heavier bumps on the tails than scale
• B(1 - W) --> prob of best and not worst; skewed, hard edges when multiply by zero. tried log-odds of this, skewed and not discriminating at top and bottom well.
• B + (1 - W) - B(1 - W) --> prob of best or not worst; skewed, hard edges, as well, at 1. tried log odds of this, doesn't discriminate at top and bottom, either.
• initializing at 1/walks instead of 0. get log odds, combine those using either std or prob means --> clump in middle, two clumps in the tails; but mean std lo has better spread, less kurtosis

its either mean(std(B), std(W -1)) or mean(std(logodds(B)), std(logodds(W) -1)) imo