Cramér's V for Categorical Association

Correlation doesn't work between categorical variables like country and plan. Cramér's V turns the chi-square statistic of their contingency table into a 0–1 strength of association:

V = \sqrt{\frac{\chi^2 / n}{\min(r - 1,\; c - 1)}}

Implement:

contingency(a, b): the table of counts as a 2-D NumPy array, with rows for the sorted unique values of a and columns for those of b
chi_square(table): $\sum \frac{(O - E)^2}{E}$ with $E_{ij} = \frac{\text{row}_i \cdot \text{col}_j}{n}$
cramers_v(a, b): 0.0 if either variable has a single category
association_matrix(df, columns): a symmetric DataFrame of pairwise V values (1.0 on the diagonal)