Some dissonant and non-dissonant notes were correctly labelled in the past, but are now being incorrectly labelled. About 1% of the notes have been affected. This issue crept in somehow when we moved to pandas 0.18.1, which no longer allowed for the forcing of string data type. Sorting out this data type issue was not so difficult, but somehow the accuracy fell a bit. I must admit that the dissonance indexer is rather clunky and in need of a rewrite. I tried to fix this issue and regain our original dissonance detection accuracy but I eventually decided that it was too time consuming of a task. Time on the dissonance indexer would be better spent rewriting it from scratch.
For VIS 3.0, I decided to accept this minor decline in dissonance classification, with the hope of being able to do this rewrite in VIS 3.1 or at least start it.
Some dissonant and non-dissonant notes were correctly labelled in the past, but are now being incorrectly labelled. About 1% of the notes have been affected. This issue crept in somehow when we moved to pandas 0.18.1, which no longer allowed for the forcing of string data type. Sorting out this data type issue was not so difficult, but somehow the accuracy fell a bit. I must admit that the dissonance indexer is rather clunky and in need of a rewrite. I tried to fix this issue and regain our original dissonance detection accuracy but I eventually decided that it was too time consuming of a task. Time on the dissonance indexer would be better spent rewriting it from scratch. For VIS 3.0, I decided to accept this minor decline in dissonance classification, with the hope of being able to do this rewrite in VIS 3.1 or at least start it.