MANUALLY EXAMINING JSYMBOLIC'S FEATURE OUTPUT IN A TEXT EDITOR

• In practice, feature values are often processed automatically with data analysis or machine learning software
  • However, it is also important to sometimes look directly at features
  • This is where we can best use our musical expertise and intuition
• However, it is useful to get a sense of what they look like
• Use your operating system's graphical (or terminal) file navigation functionality to open the "01_Comparing_Two_Files" folder you loaded the two MIDI files from earlier
• Notice that the feature extraction you just performed created four new files
  • The ACE XML feature definitions file
  • The ACE XML feature values file
  • The Weka ARFF feature values file
  • The CSV feature values file
• Using your favorite text editor, open the "MySimpleFeatureDefinitions.xml" file
  • e.g. on a Macintosh computer, right click the file > Open with > BBEdit (or whatever other text editor you have)
  • e.g. on a Windows computer, right click the file > Open with > EditPlus (or whatever other text editor you have)
  • It contains metadata about the three features you selected
• Using your favorite text editor, open the "MyJosqOckSimpleFeatureValues.xml" file
  • It contains the three extracted features for each of the two chosen files in jSymbolic's native ACE XML format
• Using your favorite text editor, open the "MyJosqOckSimpleFeatureValues.arff" file
  • It contains the three extracted features for each of the two chosen files in Weka's native ARFF format
• Using your favorite spreadsheet, open the "MyJosqOckSimpleFeatureValues.csv" file
  • It contains the three extracted features for each of the two chosen files in the standard CSV format
  • This is often the best format to use for manual analysis
• You can compare what you got with the model files pre-generated in the "Model Demo Files" folder

MANUALLY EXAMINING JSYMBOLIC'S FEATURE OUTPUT USING SPREADSHEET SOFTWARE

• It can also be useful to manually analyze extracted feature values in spreadsheet software like Microsoft Excel
  • Most spreadsheets can open CSV files, including the ones generated by jSymbolic
• As a sample of how this can be done, open the "BigJosqOckFeatureValuesManualAnalysis.xlsx" file found in the "01_Comparing_Two_Files" folder you were just working in
  • This is a file that was constructed from a jSymbolic CSV file that you will generate yourself in a later stage of this tutorial
  • In brief, this file holds the values of many features extracted by jSymbolic from a large set of pieces by Josquin and Ockeghem
  • You will need to have Microsoft Excel installed on your computer to open this file (or some other spreadsheet capable of parsing Microsoft Excel files)
• A range of simple Excel formulae have been manually added to this Excel file to enable you to see the mean, standard deviation and other statistics for each feature
  • Overall, just for Josquin or just for Ockeghem
• Look for example, at the "Vertical_Sixths" feature
  • Note how Ockeghem tends to have more vertical sixths than Josquin
  • In general, but not necessarily in any given piece
• Spreadsheets and other statistical analysis software can be used to perform these and many more sophisticated types of statistical analysis on the CSV files produced by jSymbolic
  • The rest of this tutorial will focus on using Weka (a machine learning and data mining tool) to analyze jSymbolic's feature data
  • Researchers may certainly wish to use their own alternative spreadsheets or other data analysis tools, of course

More on this sort of thing later. Next we will look at how jSymbolic's configuration files can be helpful . . .

-top of page-