Fréchet mean of rotations near identity should be near identity.

With all 3 components, inverse_transform should nearly reconstruct input.

https://github.com/MaxHalford/prince/issues/131#issuecomment-1591426031 >>> df = pd.DataFrame({ ... "foo": [1, 2, 3, 3, 5], ...

https://github.com/MaxHalford/prince/issues/161 >>> import io >>> data = pd.read_csv(io.StringIO(issue_161_data), index_col=0) >>>

https://github.com/MaxHalford/prince/issues/169 >>> import pandas as pd >>> from prince import FAMD >>> df = pd.DataFrame({'var1':[

https://github.com/MaxHalford/prince/issues/171 >>> from sklearn import impute >>> from sklearn import pipeline >>> rng = np.rando

A group made of string columns should fit (no longer raises NotImplementedError).

Non-subset Benzécri/Greenacre corrections match R's ``ca::mjca(lambda='adjusted')``. FactoMineR doesn't expose these corrections, but Greenacre's

Partial-axis correlations should match FactoMineR for the active groups. FactoMineR's ``mfa$partial.axes$cor`` includes both active and suppl

`plot` should not require rows to be drawn in order to draw columns. The column coordinates are scaled into the row coordinate space, so the row

Column coordinates for the numerical variables should match FactoMineR's quanti.var$coord.

Table 2 from Wikipedia: pairwise associations between all variables. R² for quanti-quanti, η² for quanti-quali, ϕ² (= χ²/n) for quali-quali.

Figure 1: individual factor map. FactoMineR: famd$ind$coord (signs may differ per component).

Subset-MCA Greenacre correction matches R's ``ca::mjca(lambda='adjusted', subsetcat=...)``. Background: https://github.com/MaxHalford/prince/issu

Checks that the type of the columns doesn't affect the result.

>>> df = pd.DataFrame({ ... "foo": [1, 2, 3, 3, 5], ... "bar": ["a", "b", "c", "b", "e"], ... }) >>> mca = prince.MCA(n_

Total inertia from PCA in the tangent space.

Align X to the reference shape. Parameters: X (array-like of shape (n_shapes, n_points, n_dim)): Matrix of shapes