kaggleBaseCase

jpOmaha
11/22/2014

What Determines Your KAGGLE Ranking?

  • A website for data science competitions is www.kaggle.com
  • It hosts open competitions with prizes based on final rankings
  • It also lists cumulative competitor ratings over a moving window
  • These ratings are rankings based on your competition scores

Kaggle Competitor Score:

\( scoreX = {(base/teamSizeX) * (teamRankX^{rankExpo}) * log(teams, 10) * ((window - lapse) / window)} \)

BaseCase

Team => solo; Finish => 1st; Deadline => today

base <- 100000 #Constant starting score, fixed by Kaggle
window <- 2 #Sliding window, in years
teams <- 10 # Variable
teamSizeX <- 1 # Variable
teamRankX <- 1 # Variable
lapse <- 0 # Variable
rankExpo <- -0.75 #Fixed by Kaggle
options(scipen=999) # To avoid scientific notation
(scoreX <- (base/teamSizeX) * (teamRankX^rankExpo) * log(teams, 10) * ((window - lapse) / window))
[1] 100000

After setting cor(scoreX, teamSizeX) = -1.0, scoreX = Nonlinear function of teamSizeX

plot of chunk unnamed-chunk-2

After setting cor(scoreX, teamSizeX) = -1.0, scoreX = Nonlinear function of teamRankX

plot of chunk unnamed-chunk-3