EXTRACTING A LARGER NUMBER OF FEATURES

• Let's now extract many more features, but still just for the two files we have been working with so far, using the methodology we just covered in the earlier steps of this tutorial
• Click the Select All Features button, and then deselect the final two features in order to set all features for extraction except the MEI-only features (i.e. the last two)
• Change the output file names (in the same directory as before) to "MyJosqOckManyFeatureValues.xml" and "MyManyFeatureDefinitions.xml"
• Extract the features
• Open the resulting "MyJosqOckManyFeatureValues.csv" file to take a look at the many features extracted
• Not sure what the meaning of a given feature is?
• Check the jSymbolic manual, check the feature definitions file, or double click the feature in the GUI
• Although comparing just two pieces of music can certainly be very useful and important, it can be even more revealing to compare features across many pieces of music
  • This is what we will do next

EXTRACTING FEATURES FROM A LARGER NUMBER OF FILES

• This next part will involve extracting features from a collection of many files by Josquin and Ockeghem
  • These include all the Ockeghem files in the Josquin Research Project (JRP), and all the Josquin files with the highest two Rodin attribution certainty levels
• You will be using some of the techniques covered in the earlier sections of this tutorial
  • Refer back to them if, for example, you forget how to load configuration files or to set files to be extracted with jSymbolic
• Quit and restart jSymbolic
• Load the configuration settings from the "FeaturesThatAvoidBiasInRenMusictConfigs.txt" file in the "02_Exploring_A_Larger_Dataset" sub-folder of the "jSymbolicTutorialWorkingFiles" folder you extracted onto your desktop earlier
  • This configuration file omits all features that are not relevant to this kind of Renaissance music, and which could potentially introduce problems due to systematic bias in the encoding source (as opposed to in the music itself)
    • For example, features associated with dynamics, tempo, microtones and instrumentation are all left out
  • Note that these configurations do not specify any particular input files, unlike the configuration file we saved earlier
• In the jSybmolic GUI, add all 229 of the files in the "JosquinVsOckeghem" sub-folder of the "02_Exploring_A_Larger_Dataset" folder to be extracted jSymbolic
• Set the feature values and definitions to be saved in the "02_Exploring_A_Larger_Dataset" folder
  • Call them "BigJosqOckFeatureValues.xml" and "BigJosqOckFeatureDefinitions.xml", respectively
  • Also leave jSymbolic set to generate CSV and ARFF files
• Extract the features
  • This may take a little while, so we can begin to look at Weka in the meantime

On to Weka . . .

-top of page-