haydn-mozart

Classification of Haydn and Mozart String Quartets

This repository provides the datasets and code used to classify string quartet movements as being composed by either Haydn or Mozart.

Datasets

The data consist of 292 **kern files obtained from the KernScores website http://kern.ccarh.org/ (maintained by the Center for Computer Assisted Research in the Humanities at Stanford University) and 107 additional **kern files kindly provided by Dr. Peter van Kranenberg. They are organized into two datasets:

• HM107: 54 Haydn movements and 53 Mozart movements, as used in the van Kranenburg and Backer (2005) and Velarde et al. (2018, 2016) studies.
• HM285: 203 Haydn movements and 82 Mozart movements.

Code

The code to produce the analyses in our paper are provided as scripts written in the statistical programming language R. Please set your R working directory to the location of your local repository before running the scripts.

A brief description of each script follows, and further documentation is provided within the script itself.

Main analyses

• 1_create_feature_sets.R: reads in the **kern files, cleans the data, and creates the feature sets
• 2_classify_HM285.R: classifies the composer of each movement in the HM285 dataset using the leave-one-out (LOO) CV scheme
• 3_classify_HM107.R: classifies the composer of each movement in the HM107 dataset using the LOO CV scheme
• 4_classify_simple_HM285.R: classifies the composer of each movement in the HM285 dataset using the LOO CV scheme with a set of simple features only
• 5_classify_LOQO_HM285.R: classifies the composer of each movement in the HM285 dataset using the leave-one-quartet-out (LOQO) CV scheme
• 6_classify_full_HM285.R: fits one model on all 285 movements in the HM285 dataset for interpretation purposes
• 7_interpret_results.R: produces some plots, tables, and other summaries of results on the HM107 and HM285 datasets

Supplementary analyses

• S1-1_create_expanded_feature_sets.R: creates expanded feature sets that includes all segment lengths for the segment features
• S1-2_analyze_segment_features.R: makes heatmaps of correlation matrices for sets of segment features for assessing similarity
• S2_make_feature_heatmaps.R: generates heatmaps of the feature matrix X, the Gram matrix XX', and their variants
• S3_analyze_predictions.R: analyzes the estimated probabilities, classes, and selected features from LOO, LOQO, and LOO on the reduced/simple feature set

Reference

For a full discussion of our methods, please see our paper "Where Does Haydn End and Mozart Begin? Composer Classification of String Quartets" (https://arxiv.org/abs/1809.05075).