In this section, we will use an EM implementation to fit a Gaussian mixture model with diagonal covariances to a subset of the Wikipedia dataset.

• .zip file is data file.
  • people_wiki.gl.zip (unzip people_wiki.gl) consists of 59,071 samples and 3 features.
• description files
  • .ipynb file is the solution of Week 4 program assignment 2
    • 4_em-with-text-data.ipynb
  • .html file is the html version of .ipynb file.
    • 4_em-with-text-data.html
  • .py
    • 4_em-with-text-data.py
  • file
    • 4_em-with-text-data

• Recommend open md file inside a file
• open .html file via brower for quick look.

• EM algorithm.

• Load Wikipedia data and extract TF-IDF features
• EM in high dimensions
  • Log probability function for diagonal covariance Gaussian.
  • EM algorithm for sparse data.
    • Initializing mean parameters using k-means.
    • Initializing cluster weights.
    • Initializing covariances
    • Running EM
• Interpret clusters
• Comparing to random initialization