Welcome to the main website of ppidyom!
What is ppidyom?
ppidyom is an R package for applying partial predictive matching (PPM) algorithms to musical data. PPM algorithms are a sophisticated form of N-gram model, where musical events (notes, chords, etc.) are predicted based on the previous music. This (probabilistic) predictions can then be used as the basis for information theory metrics, like information content and entropy.
ppidyom is similar to the widely used IDyOM (Pearce, 2005) and ppm (Harrison, et al. 2020) softwares. The goal of the ppidyom project is to make partial-predictive-modeling faster, easier, and more transparent than these existing systems. In particular, ppidyom is designed to work within the humdrumR package framework, making a complete system for musicological analysis.
What is partial predictive matching?
PPM models originated in the field of computational linguistics. PPM models are a form of N-gram model, where each sequential musical event (note, chord, etc.) is predicted based on the N-previous events. In a PPM model, N-grams of various lengths are “blended” to create the final prediction.
PPM model implementations can also incorporate:
- Long-term and short-term models.