Accessing the values of correlation matrices

In some cases, directly accessing the values underneath the default visualizations provided in the report (like correlation matrices) may be useful - to generate a highly custom visualization, to reuse the computations, to do more follow-up computations, to directly change the values according to some domain-specific logic. This is possible by exploring the deeper structure of the ProfileReport object.

The snippet below shows how to list the available correlation matrices:

Listing available correlation information in the report

  # Computing the report
  data = pd.read_csv(
      "https://raw.githubusercontent.com/datasciencedojo/datasets/master/titanic.csv"
  )
  profile = ProfileReport(data, title="Titanic")

  # Listing available correlation
  correlations = profile.description_set["correlations"]
  print(correlations.keys())

In this case, all the 6 possible correlation metrics were computed and the output would be dict_keys(['auto', 'spearman', 'pearson', 'kendall', 'cramers', 'phi_k']). Each correlations[CORRELATION_TYPE] object is a pandas DataFrame. To access its values:

Accessing the values of the Pearson correlation

  # DataFrame where row and column names are the names of the original columns
  pearson_df = correlations["pearson"]

  # Actual correlation values
  pearson_mat = pearson_df.values
  print(pearson_mat)

Additional information is available on this StackOverflow answer.