Use original (or at least consistent) outcome labels in id2outcome.txt

[logological]

Currently id2outcome.txt uses numeric IDs for the classification outcomes, but (at least for cross-validation experiments) these IDs are not consistent from file to file. For example, BrownPosDemo does a two-fold cross-validation. One of the id2outcome.txt file uses the following mapping of numeric IDs to the original labels:

0=NPg 2=JJ 1=(null) 3=RB 5=TO 4=PPS 6=RP 7=NP 8=NN 10=VBN 9=VB 11=pct 12=PPO 13=BE 14=MD 15=DTS 16=VBZ 17=AT 18=IN 19=CS 20=VBG 21=VBD 22=BEDZ 23=NNS 24=CC 25=CD 26=AP 27=PPg

The other id2outcome.txt file uses a slightly different mapping:

0=NPg 2=(null) 1=JJ 3=RB 5=PPS 4=TO 6=RP 7=NP 8=NN 10=VB 9=VBN 11=pct 12=PPO 13=BE 14=MD 15=DTS 16=VBZ 17=AT 18=IN 19=CS 20=VBG 21=VBD 22=BEDZ 23=NNS 24=CC 25=CD 26=AP 27=PPg

In order to get the raw classifications, not only do I need to combine the two files, but I first have to manually un-map all the numeric IDs to their original labels.

It would be better if id2outcome.txt didn't use numeric IDs at all, but rather used the original label IDs. If for some reason the mapping to numeric IDs is necessary, it would be helpful if the mapping were consistent across files.

[daxenberger]

InnerBatchUsingTCEvaluationReport now generates a file "id2harmonizedOutcome.txt" in the crossvalidation context folder.