Hands-On Unsupervised Learning with Python

Contingency matrix

A very simple and powerful tool that can show the performance of a clustering algorithm when the ground truth is known is the contingency matrix C_m. If there are m classes, C_m ∈ ℜ^{m × m} and each element C_m(i, j) represents the number of samples with Y_true = i that have been assigned to the cluster j. Hence, a perfect contingency matrix is diagonal, while the presence of elements in all the other cells indicates a clustering error.

In our case, we obtain the following:

from sklearn.metrics.cluster import contingency_matrixcm = contingency_matrix(kmdff['diagnosis'].apply(lambda x: 0 if x == 'B' else 1), kmdff['prediction'])

The output of the previous snippet can be visualized as a heat map (the variable cm is a (2 × 2) matrix): ...