What are the minimum characteristics that a binary matrix must meet to avoid the following error: "Insufficient Number of data since union is 0"?

[Galy88]

Expected Behavior

Given a binary matrix, for example, of a 3x4 size, the KModes algorithm is expected to run and find to which cluster each row of the matrix should belong.

Actual Behavior

Currently we have a 3x4 binary matrix, when we run the KModes algorithm, we immediately get the following message: "Insufficient Number of data since union is 0". With 3x4 binary matrixes sometimes the algorithm works.

Steps to Reproduce the Problem

1. Import libraries:

   import numpy as np
   from kmodes.kmodes import KModes
   from kmodes.util.dissim import jaccard_dissim_binary

   2. Create binary matrix:

      m = np.array([[0, 1, 1, 1],
            [0, 1, 1, 0],
            [1, 1, 1, 0]])

2. Adjust kmodes km = KModes(n_clusters=2, init='cao', random_state=0, n_jobs=-1, cat_dissim=jaccard_dissim_binary)

3. Fit predict clusters = km.fit_predict(m)

with this matrix the algorithm works:

m2 = np.array([[1, 0, 0, 1], 
               [0, 1, 1, 0],
               [0, 1, 1, 0]])

clusters = km.fit_predict(m2)

Specifications

• kmodes 0.12.1
• Python 3.7.13

[nicodv]

This appears to occur due to the jaccard_dissim_binary dissimilarity function, in combination with some specifics of the data set. Specifically, it appears this distance metric does not support situations where you have binary data, and all rows have the same value for a column.

You'll either need to:

• Add some more variety to your data.
• Use a different dissimilarity metric.
• Drop constant columns from the data beforehand. Since there's no point in differentiating between such columns, you may as well. That way, you can still use this distance function.