DataFrame support

Prince accepts pandas and Polars DataFrames in every estimator that consumes two-dimensional tabular data: PCA, CA, MCA, FAMD, MFA, and PGA. Narwhals provides the DataFrame compatibility layer. Prince converts these inputs to pandas DataFrames and NumPy arrays internally, so pandas remains a required dependency.

Polars is optional. Install it alongside Prince if you want to use Polars DataFrames:

pip install prince polars

Polars input

Pass a Polars DataFrame directly to fit, transform, or fit_transform. This works with PCA, CA, MCA, FAMD, MFA, and PGA.

from __future__ import annotations

import polars as pl

import prince

X = pl.DataFrame(
    {
        "height": [1.61, 1.72, 1.80, 1.68, 1.75],
        "weight": [54.0, 68.0, 81.0, 63.0, 75.0],
        "age": [23.0, 35.0, 29.0, 41.0, 32.0],
        "score": [7.2, 8.1, 8.8, 7.7, 8.4],
    }
)
X

Prince returns a pandas DataFrame by default. This preserves the output behavior of earlier Prince versions. Because Polars and NumPy do not have an index, their default result uses a pandas RangeIndex. A pandas input keeps its original index.

coordinates = prince.PCA(n_components=2, random_state=42).fit_transform(X)
print(type(coordinates))
coordinates

<class 'pandas.DataFrame'>

Polars output

Use scikit-learn’s set_output API when the result should also be a Polars DataFrame.

pca = prince.PCA(n_components=2, random_state=42).set_output(transform="polars")
coordinates_polars = pca.fit_transform(X)
print(type(coordinates_polars))
coordinates_polars

<class 'polars.dataframe.frame.DataFrame'>

NumPy input

PCA also accepts a two-dimensional NumPy array. Array input has no column names, and the default pandas result uses a RangeIndex.

X_numpy = X.to_numpy()
pca = prince.PCA(n_components=2, random_state=42).fit(X_numpy)
coordinates = pca.transform(X_numpy)
print(type(coordinates))
coordinates

<class 'pandas.DataFrame'>

NumPy output

Pass as_array=True to PCA’s transform, fit_transform, or inverse_transform method to return a NumPy array instead of the default pandas DataFrame.

coordinates_numpy = pca.transform(X_numpy, as_array=True)
print(type(coordinates_numpy))
coordinates_numpy

<class 'numpy.ndarray'>





array([[ 2.85024267, -1.11448151],
       [-0.19077026,  0.48028256],
       [-2.27863386, -0.82529274],
       [ 0.74360471,  1.62134058],
       [-1.12444326, -0.1618489 ]])

Estimator coverage

Polars input is supported by PCA, CA, MCA, FAMD, MFA, and PGA. GPA is different: it consumes a three-dimensional array with shape (n_shapes, n_points, n_dimensions), which cannot be represented as a two-dimensional pandas or Polars DataFrame. GPA therefore continues to accept array-like input.