galuppi

Repository reproducing the analysis from the paper:

Llorens, A., García-Portugués, E., and Torrente, Á. (2026). By Baldassare Galuppi? Inferring the authorship of three untraceable opera seria arias. Submitted.

Data

Three corpora of eighteenth-century arias, curated at the ERC Didone project and described by their musif features, back the three classification problems of the paper:

Corpus	Arias	Problem
code/gp.csv	300	Binary, Baldassare Galuppi vs. Davide Perez; the 3 arias of unknown authorship carry NA as Composer
code/gp_multi.csv	426	Demofoonte, 5 candidate composers
code/gp_ale.csv	655	Alessandro nell'Indie, 6 candidate composers

Each file has an AriaId column, a Composer column, and the 8,244 musif features extracted from the scores. code/aria_names.csv maps every AriaId to its title and opera. The documentation of the features and of the software that extracts them is available at https://musif.didone.eu.

Reproducing the analysis

From the repository root:

quarto render code/galuppi.qmd

The knitr cache of the thirty modelling chunks is shipped in code/galuppi_cache/, so this is a warm render: it restores the fitted models instead of refitting them, and reproduces the numbers of the paper exactly.

The render writes:

Output	Contents
code/galuppi.html	Full analysis report
paper_numbers.tex	The \gp... macros the manuscript inputs
code/paper_numbers.csv	The same numbers as a table
imgs/	The three figures of the paper
code/discussion/exemplar_arias.csv, code/discussion/ale_exemplar_arias.csv, code/discussion/multi_exemplar_arias.csv	Arias exemplifying each selected stylistic marker
code/marker_correlates.csv, code/ale_marker_correlates.csv, code/multi_marker_correlates.csv	Features correlating with each selected marker
code/discussion/marginal_markers.csv, code/discussion/doubtful_arias_markers.csv	Marker summaries for the corpora and for the three doubtful arias

The folder code/discussion/ gathers the evidence behind the Discussion section of the paper, namely the exemplar arias of each stylistic marker, the marker summaries, and the two counts taken from the scores themselves along with the scripts that produce them.

code/coverage_study.R is the simulation study of the appendix, which measures the coverage of the intervals at the settings of the three problems.

Requirements

R (>= 4.4), Quarto, a Java runtime for the h2o cluster, and the R packages tidyverse, tidymodels, agua, h2o, glmnet, MASS, ks, doParallel, foreach, future, doFuture, progressr, withr and testthat; code/discussion/viola_doubling.py additionally needs Python with music21.