Project: Words 4 Music

Project Description

Term: Fall 2016

• Data link-(courseworks login required)
• Data description
• Contributor's name: Jaime Gacitua
• Project title: GBM to Recommend song lyrics
• Project summary: A model is proposed to predict song lyrics given song features. A brief description of the model is below.

---

• We have a dictionary of 5000 words and, 2700 songs.

• We have a matrix that, for every song, indicates how many times each word appears (bag of words)

• We also have access to multiple features, for every song

• The 19 features chosen to predict words are the following:

  1. tempo.median
  2. tempo.var
  3. tatums.median
  4. tatums.var
  5. loudness.median
  6. loudness.var
  7. duration
  8. timbre (median of each of the 12 dimensions)

• The word matrix is converted into a binary matrix.

  • If a song is present in a song, the value is 1. Otherwise, 0.

• For each column (word) of the matrix, a Generalized Boosting Model (GBM) was fitted, with bernoulli responses.

  • The 19 features are the input, and the (0-1) word column is the output.
  • In total 5000 GBM models are fit.

• Parameter tuning was done using cross validation.

  • The error is calculated using the sum of ranks.

• The model trains in around 30 minutes

• The best average sum of ranks achieved was 0.229, and the simplest model for that result was n=100 trees and depth=8.

Following suggestions by RICH FITZJOHN (@richfitz). This folder is orgarnized as follows.

proj/
├── lib/
├── data/
├── doc/
├── figs/
└── output/

Please see each subfolder for a README file.