New first column after clustering causes label type issues
USE CASE: WHAT DO YOU WANT TO DO?
Label types should be consistent before and after clustering
STEPS TO REPRODUCE AN ISSUE (OR TRIGGER A NEW FEATURE)
This issue has been mitigated by line 367 in ModelLoader.java. To reproduce the issue, you must comment out that line. See the COMMENT below for details on the line.
- Open small_133x133.txt
- Cluster with default options
CURRENT BEHAVIOR
If line 367 of ModelLoader is commented, then the AID row loses its ARRY#X values and replaces them with the COMPLEX values. Also see issue #492.
EXPECTED BEHAVIOR
The AID labels should be retained and assigned the correct label type after clustering.
DEVELOPERS ONLY SECTION
SUGGESTED CHANGE (Pseudocode optional)
This is happening because the new index of the label type is being used to retrieve the old labels where that index is different. The change in the indexes by adding a column from old to new should be accounted for when assigning labels.
FILES AFFECTED (where the changes will be implemented) - developers only
unknown
LEVEL OF EFFORT - developers only
minor
COMMENTS
ModelLoader.java, line 367 (inside parseLabelTypes: readColLabelTypes[i] = stringLabels[i][1];
) preceded by the comment:
//The following is a kludge because clustering
//inserts a new column at the beinning which renames
//the original first row header. This assumes that
//the first column is now the second column and that
//this will only occur on the first row. This also
//assumes that the label type is the same for the
//first row and (original) "first" column
//TODO: The bug that causes this issue must be
//fixed.