the graphs showing pretty colors.(median vs variance vs # of features)

we found that only ret2,ret3 had results > baseline
we found that only ret3 had larger results > baseline
we found that round(#attributes/20) = 40 had lowest pfs


so we filtered  ret3_5_100.csv  for #attr 25 <= A <= 45
  cat ret3_5_100.csv | gawk -F, '30 <= $2 && $2 <=50 {print}' > breadcrumbs/test

winLossTie --input breadcrumbs/test --test mw --perform 12 --fields 15 --key 2,3,5 > breadcrumbs/winLossTie.txt

so we mann-whitney those results

ound all results with no losses, sorted on wins
 breadcrumbs/winLossTieNoLosses.txt

find the number of wins, graph them. 

find the group where you win at least half as much as you tie


then fr the Y's look at t pd pd (see breadcrumbs/pdf.txt)

final: plot.