bliss-rs/src/distance.rs
2021-08-23 18:16:32 +02:00

240 lines
8.3 KiB
Rust

//! Module containing various distance metric functions.
//!
//! All of these functions are intended to be used with the
//! [custom_distance](Song::custom_distance) method, or with
//! [playlist_from_songs_custom_distance](Library::playlist_from_song_custom_distance).
//!
//! They will yield different styles of playlists, so don't hesitate to
//! experiment with them if the default (euclidean distance for now) doesn't
//! suit you.
#[cfg(doc)]
use crate::Library;
use crate::Song;
use crate::NUMBER_FEATURES;
use ndarray::{Array, Array1};
use ndarray_stats::QuantileExt;
use noisy_float::prelude::*;
/// Convenience trait for user-defined distance metrics.
pub trait DistanceMetric: Fn(&Array1<f32>, &Array1<f32>) -> f32 {}
impl<F> DistanceMetric for F where F: Fn(&Array1<f32>, &Array1<f32>) -> f32 {}
/// Return the [euclidean
/// distance](https://en.wikipedia.org/wiki/Euclidean_distance#Higher_dimensions)
/// between two vectors.
pub fn euclidean_distance(a: &Array1<f32>, b: &Array1<f32>) -> f32 {
// Could be any square symmetric positive semi-definite matrix;
// just no metric learning has been done yet.
// See https://lelele.io/thesis.pdf chapter 4.
let m = Array::eye(NUMBER_FEATURES);
(a - b).dot(&m).dot(&(a - b)).sqrt()
}
/// Return the [cosine
/// distance](https://en.wikipedia.org/wiki/Cosine_similarity#Angular_distance_and_similarity)
/// between two vectors.
pub fn cosine_distance(a: &Array1<f32>, b: &Array1<f32>) -> f32 {
let similarity = a.dot(b) / (a.dot(a).sqrt() * b.dot(b).sqrt());
1. - similarity
}
/// Sort `songs` in place by putting songs close to `first_song` first
/// using the `distance` metric.
pub fn closest_to_first_song(
first_song: &Song,
songs: &mut Vec<Song>,
distance: impl DistanceMetric,
) {
songs.sort_by_cached_key(|song| n32(first_song.custom_distance(song, &distance)));
}
/// Sort `songs` in place using the `distance` metric and ordering by
/// the smallest distance between each song.
///
/// If the generated playlist is `[song1, song2, song3, song4]`, it means
/// song2 is closest to song1, song3 is closest to song2, and song4 is closest
/// to song3.
///
/// Note that this has a tendency to go from one style to the other very fast,
/// and it can be slow on big libraries.
pub fn song_to_song(first_song: &Song, songs: &mut Vec<Song>, distance: impl DistanceMetric) {
let mut new_songs = Vec::with_capacity(songs.len());
let mut song = first_song.to_owned();
while !songs.is_empty() {
let distances: Array1<f32> =
Array::from_shape_fn(songs.len(), |i| song.custom_distance(&songs[i], &distance));
let idx = distances.argmin().unwrap();
song = songs[idx].to_owned();
new_songs.push(song.to_owned());
songs.retain(|s| s != &song);
}
*songs = new_songs;
}
#[cfg(test)]
mod test {
use super::*;
use crate::Analysis;
use ndarray::arr1;
use std::path::Path;
#[test]
fn test_song_to_song() {
let first_song = Song {
path: Path::new("path-to-first").to_path_buf(),
analysis: Analysis::new([
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
]),
..Default::default()
};
let first_song_dupe = Song {
path: Path::new("path-to-dupe").to_path_buf(),
analysis: Analysis::new([
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
]),
..Default::default()
};
let second_song = Song {
path: Path::new("path-to-second").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 1.9, 1., 1., 1.,
]),
..Default::default()
};
let third_song = Song {
path: Path::new("path-to-third").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2.5, 1., 1., 1.,
]),
..Default::default()
};
let fourth_song = Song {
path: Path::new("path-to-fourth").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 0., 1., 1., 1.,
]),
..Default::default()
};
let mut songs = vec![
first_song.to_owned(),
first_song_dupe.to_owned(),
second_song.to_owned(),
third_song.to_owned(),
fourth_song.to_owned(),
];
song_to_song(&first_song, &mut songs, euclidean_distance);
assert_eq!(
songs,
vec![
first_song,
first_song_dupe.to_owned(),
second_song,
third_song,
fourth_song
],
);
}
#[test]
fn test_sort_closest_to_first_song() {
let first_song = Song {
path: Path::new("path-to-first").to_path_buf(),
analysis: Analysis::new([
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
]),
..Default::default()
};
let first_song_dupe = Song {
path: Path::new("path-to-dupe").to_path_buf(),
analysis: Analysis::new([
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
]),
..Default::default()
};
let second_song = Song {
path: Path::new("path-to-second").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 1.9, 1., 1., 1.,
]),
..Default::default()
};
let third_song = Song {
path: Path::new("path-to-third").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2.5, 1., 1., 1.,
]),
..Default::default()
};
let fourth_song = Song {
path: Path::new("path-to-fourth").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 0., 1., 1., 1.,
]),
..Default::default()
};
let fifth_song = Song {
path: Path::new("path-to-fifth").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 0., 1., 1., 1.,
]),
..Default::default()
};
let mut songs = vec![
first_song.to_owned(),
first_song_dupe.to_owned(),
second_song.to_owned(),
third_song.to_owned(),
fourth_song.to_owned(),
fifth_song.to_owned(),
];
closest_to_first_song(&first_song, &mut songs, euclidean_distance);
assert_eq!(
songs,
vec![
first_song,
first_song_dupe,
second_song,
fourth_song,
fifth_song,
third_song
],
);
}
#[test]
fn test_euclidean_distance() {
let a = arr1(&[
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 0.,
]);
let b = arr1(&[
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0.,
]);
assert_eq!(euclidean_distance(&a, &b), 4.242640687119285);
let a = arr1(&[0.5; 20]);
let b = arr1(&[0.5; 20]);
assert_eq!(euclidean_distance(&a, &b), 0.);
assert_eq!(euclidean_distance(&a, &b), 0.);
}
#[test]
fn test_cosine_distance() {
let a = arr1(&[
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 0.,
]);
let b = arr1(&[
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0.,
]);
assert_eq!(cosine_distance(&a, &b), 0.7705842661294382);
let a = arr1(&[0.5; 20]);
let b = arr1(&[0.5; 20]);
assert_eq!(cosine_distance(&a, &b), 0.);
assert_eq!(cosine_distance(&a, &b), 0.);
}
}