test 6 5 ideal

[tdhock]

with old code (key by freq) we have

Key: <g.int, fold.out>
   g.int fold.out    s1    s2  f1s1  f1s2  f2s1  f2s2
   <int>    <int> <int> <int> <num> <num> <num> <num>
1:     1        1     5     3     5     3     0     0
2:     2        2     1     2     5     3     1     2
3:     3        2     1     2     5     3     2     4
4:     4        2     1     2     5     3     3     6
5:     5        2     3     0     5     3     6     6
6:     6        1     1     1     6     4     6     6

above is bad because ideal is 6 5 but we get 6 4 and 6 6.

• 1 2 has same RSS and number of samples as 3 0
• all three 1 2 groups go into fold 2, putting it at 3 6, already over ideal of 5 before 3 0 and 1 1 are considered.

below is with new cod, good because 3 0 group is considered before 1 2 groups.

   g.int fold.out    s1    s2  f1s1  f1s2  f2s1  f2s2
   <int>    <int> <int> <int> <num> <num> <num> <num>
1:     1        1     5     3     5     3     0     0
2:     2        2     3     0     5     3     3     0
3:     3        2     1     2     5     3     4     2
4:     4        2     1     2     5     3     5     4
5:     5        1     1     2     6     5     5     4
6:     6        2     1     1     6     5     6     5

this is a simple example which can be used to justify the neg_freq sort heuristic. (it is also more consistent with the RSS minimization heuristic, which prioritizes groups with counts in more frequent classes)

[tdhock]

Preview