#2 Wrong Feature Ids in ConcatRegexFeatures

open
nobody
None
5
2006-07-09
2006-07-09
Amit Jaiswal
No

ConcatRegexFeatures is giving warning as same name is
used for different feature ids which is an indicator of
some problem in feature id generation.

Here is how to reproduce it.
addFeature(new FeatureTypesEachLabel(this, new
ConcatRegexFeatures(this, -1, 1))))

Basically set the relSegmentStart and relSegmentEnd to
(-1, 1) and run iitb.Segment.Segment package.

Looks like idBase variable is set to 2^(window-1)
instead of 2^window in ConcatRegexFeatures.
-amit

Discussion

  • Logged In: NO

    Yes, Amit is right. The idBase should be 2^window and not
    2^window-1. Infact, in the earliar code, it was 2^window
    only. This way, for each pattern index in the feature types,
    there is a 2^window (0 to (2^window-1)) feature ids reserved
    at different offsets for each pattern index. If we keep it
    2^(window-1), then feature ids will overlap as there is no
    sufficient gap between two pattern's baseId.