Changelog

0.19.0

MFA: column_cosine_similarities_ was not callable. Calling mfa.column_cosine_similarities_(X) raised TypeError: 'DataFrame' object is not callable because a method override was trying to call the parent class's @property. The override was removed — the inherited property works correctly. Reported in #218.
MFA: partial_row_coordinates broken with supplementary groups. The method crashed with a shape mismatch when supplementary groups were specified, because column weights and the data matrix had incompatible dimensions.
MFA: partial_row_coordinates used wrong scaling with supplementary rows. The method recomputed mean/std from the input data (including supplementary rows) instead of using the fitted scaler parameters, producing incorrect results.

MFA: group-level results. Added group_coordinates_, group_contributions_, and group_cosine_similarities_ properties. These summarize how each group of variables relates to the global MFA components, corresponding to FactoMineR's result$group$coord, result$group$contrib, and result$group$cos2. Requested in #217.
MFA: partial axes results. Added partial_correlations_ and partial_contributions_ properties. These describe how each group's own PCA axes relate to the global MFA axes, corresponding to FactoMineR's result$partial.axes$cor and result$partial.axes$contrib. Requested in #217.
MFA: partial axes correlation plot. Added plot_partial method. Produces a correlation circle showing how each group's PCA axes project onto the global MFA plane, corresponding to FactoMineR's plot.MFA(res, choix="axes").

FAMD: fix categorical variable normalization. The proportions used to normalize one-hot encoded columns were computed as n_k / (n * num_cat_vars) * 2 instead of n_k / n. This only produced correct results when the dataset had exactly 2 categorical variables. Reported in #215.

FAMD: column_correlations property. Returns the signed Pearson correlations between standardized quantitative variables and principal components (corresponding to FactoMineR's quanti.var$coord). These are the values plotted in a correlation circle. For qualitative variables, η² values are returned (same as column_coordinates_).