A sphere of fifths

3d plot of chord space

from diffusers import DiffusionPipeline
from transformers import set_seed
from PIL import Image
import torch
import random
import ssl
import os
ssl._create_default_https_context = ssl._create_unverified_context

#locate library
#model_id = "./stable-diffusion-v1-5"
model_id = "dreamshaper-xl-turbo"

pipeline = DiffusionPipeline.from_pretrained(
    pretrained_model_name_or_path = "../../../../bigFiles/huggingface/dreamshaper-xl-turbo/"
)

pipeline = pipeline.to("mps")

# Recommended if your computer has < 64 GB of RAM
pipeline.enable_attention_slicing("max")

prompt = "piano chord. 3d sphere. piano sphere. retro. 80s cartoon."

for s in range(30):
  for n in [5,10]:
    seed = s+21
    num_steps = n+1
    set_seed(seed)
    
    image = pipeline(prompt,height = 1024,width = 1024,num_images_per_prompt = 1,num_inference_steps=num_steps)
    
    image_name = "images/synth_{}_{}.jpeg"
    
    image_save = image.images[0].save(image_name.format(seed,num_steps))

piano chord. 3d sphere. piano sphere. retro. 80s cartoon. - Dreamshaper v7

My previous graphs of chord space have been 2-dimensional. I haven’t looked at the eigendecomposition of the similarity matrix, so I’m not sure how many “meaningful” dimensions there are here. Nevertheless, there are more dimensions than implied by the 2d graph. I’ve been meaning to use MDS to create a 3d plot, and that’s what I’m doing here, hopefully with plotly so it can be spun around interactively in the browser.

Chord sphere

library(tidyverse)
# pre-processing to get the chord vectors

# load chord vectors
c_chord_excel <- rio::import("chord_vectors.xlsx")

# grab feature vectors
c_chord_matrix <- as.matrix(c_chord_excel[,4:15])

# assign row names to the third column containing chord names
row.names(c_chord_matrix) <- c_chord_excel[,3]

# define all keys
keys <- c("C","Db","D","Eb","E","F","Gb","G","Ab","A","Bb","B")


# the excel sheet only has chords in C
# loop through the keys, permute the matrix to get the chords in the next key
# add the permuted matrix to new rows in the overall chord_matrix
for (i in 1:length(keys)) {

  if (i == 1) {
    # initialize chord_matrix with C matrix
    chord_matrix <- c_chord_matrix

  } else {
    #permute the matrix as a function of iterator
    new_matrix <- cbind(c_chord_matrix[, (14-i):12],c_chord_matrix[, 1:(13-i)] )

    # rename the rows with the new key
    new_names <- gsub("C", keys[i], c_chord_excel[,3])
    row.names(new_matrix) <- new_names

    # append the new_matrix to chord_matrix
    chord_matrix <- rbind(chord_matrix,new_matrix)

  }
}

chord_properties <- tibble(
  type = rep(c_chord_excel$type,length(keys)),
  key = rep(keys, each = dim(c_chord_matrix)[1]),
  chord_names  = row.names(chord_matrix),
  synonyms = list(NA),
  database_chord = FALSE
)

first_order <- lsa::cosine(t(chord_matrix))

# find repeats and build synonym list
repeat_indices <- c()

first_occurrence <- c()

for(i in 1:dim(chord_matrix)[1]){
  # get the current row
  evaluate_row <- first_order[i,]

  # don't count the current item as a repeat
  evaluate_row[i] <- 0

  # repeats are the ids for any other 1s found
  repeats <- which(evaluate_row == 1 )

  if(length(repeats) == 0){
  }

  if(length(repeats) > 0){
    #add to list of repeat items
    repeat_indices <- c(repeat_indices,repeats)

    # add synonyms
    chord_properties$synonyms[i] <- list(synonyms = row.names(chord_matrix)[repeats])
  }

  if(i %in% first_occurrence == FALSE){
    if(i %in% repeat_indices == FALSE){
      first_occurrence <- c(first_occurrence,i)
      chord_properties$database_chord[i] <- TRUE
    }
  }
}

chord_properties <- chord_properties %>%
  mutate(num_notes = rowSums(chord_matrix),
         id = 1:dim(chord_matrix)[1])

# keep only unique chord, recompute similarities
chord_matrix_no_repeats <- chord_matrix[first_occurrence,]
first_order_no_repeats <- lsa::cosine(t(chord_matrix_no_repeats))
second_order_no_repeats <- lsa::cosine(first_order_no_repeats)

# remove scales and individual notes
only_chords <- chord_properties %>%
  filter(type!="scale",
         type!="key",
         database_chord == TRUE)

first_order_chords <- first_order_no_repeats[only_chords$chord_names,
                                               only_chords$chord_names]
second_order_chords <- second_order_no_repeats[only_chords$chord_names,
                                               only_chords$chord_names]

library(ggplot2)
library(ggrepel)
library(plotly)
library(bslib)

mds_first_order <- cmdscale((first_order-1),k=3)
mds_first_order <- as_tibble(mds_first_order) %>%
  cbind(chord_properties) %>%
  mutate(bold_me = case_when(type == "key" ~ 8,
                             type != "key" ~ 7)) %>%
  filter(type %in% c("scale","other") == FALSE,
         num_notes <= 4)

fig <- plot_ly(mds_first_order, x = ~V1, y = ~V2, z = ~V3, color = ~type, mode='text', text = ~chord_names)

fig |>
  bslib::card(full_screen = TRUE)

I took out a bunch of chords so this was easier to look at. The plotly graph should go full screen and has some controls for spinning the sphere around and zooming in etc. If you click the names in the legend you can hide or show the corresponding chords, which can make it easier to see how things shake out.

In terms of the circle of fifths, the new z dimension puts 6 of them up top, and 6 of them down below. Within the X-Y plane, there are two alternating circles of fifths (CDEGbAbBb, and FGABDbEb).

Neat to poke around here.