Lodestar.Preprocessing 0.2.0. This page is frozen at that release. Read the current documentation for what main says now. A link to a decision or a migration page follows main, and leaves the archive.

Home › Feature transforming

PolynomialFeaturesOptions

What PolynomialFeatures expands, and how far.

public sealed record PolynomialFeaturesOptions

Properties — Degree is the highest total power a term may reach; scikit-learn's degree, default 2. InteractionOnly keeps only terms in which every feature appears at most once; scikit-learn's interaction_only, default false. IncludeBias emits the constant term; scikit-learn's include_bias, default true.

Example — interactions without powers and without a bias.

using Lodestar.Preprocessing;

var options = new PolynomialFeaturesOptions
{
    Degree = 2,
    InteractionOnly = true,
    IncludeBias = false,
};

string[] names = PolynomialFeatures.FeatureNames(3, options);

string joined = string.Join("|", names);   // => x0|x1|x2|x0 x1|x0 x2|x1 x2

Remarks — degree 0 with the bias off is refused, on all three members: the pair leaves no term, and the reference raises rather than returning a matrix with no columns.

order is absent. scikit-learn's third parameter chooses numpy's memory layout, C-contiguous or Fortran-contiguous, which changes how the array is stored and not what it holds. There is one layout here, and it is the row-major one every member of this package uses.

Being a record, two option sets with the same three values are equal.

Applies to — net10.0, netstandard2.0.

See also — PolynomialFeatures.Transform, the Python equivalence table.