jSymbolic Tutorial

jSymbolic Tutorial - Portuguese Research

LOOKING AT DATA FROM THE THE ANATOMY OF LATE 15TH- AND EARLY 16TH-CENTURY IBERIAN POLYPHONIC MUSIC PROJECT

Dr. João Pedro d’Alvarenga has kindly provided samples of the music associated with a very interesting project his research group is working on
- This music is found in the "04_Iberian_Data" folder in the "jSymbolicTutorialWorkingFiles.zip" file you downloaded earlier
This data provides an excellent opportunity to review what we have learned and to explore data that is particularly relevant to the context of this tutorial
It is suggested that you once again use the provided "FeaturesThatAvoidBiasInRenMusictConfigs.txt" feature subset
Please load some of these files into jSymbolic and run a feature extraction
- You may wish to start with a subset of the files, depending on the time available
Once feature extraction is complete, please explore the extracted features in Weka
- But do not forget to label the music using a spreadsheet first!
  - And you may need to be careful about certain non-ASCII characters in the class labels, as Weka can be fussy about some of these
Things to experiment with in Weka:

Manual examination of feature scatterplots in Weka
Automatic feature selection
Machine learning
- Weka can use the jSymbolic features to distinguish between the four groups of music about 85% of the time under cross-validation, with just default settings
  - It does tend to confuse the Penalosa (secular) as Penalosa (sacred), however, which is actually a musicologically reasonable result
- You can also try separating out your own training and testing sets, and seeing how individual pieces are actually classified
  - Try to keep an equally balanced level of music in each category in the training set and in the testing set

What would you like to know about this music?
- What kinds of things would you like the features to reveal?

A few final words remain to be said about jSymbolic . . .