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

BinEncoding

What a row transformed by KBinsDiscretizer carries.

public enum BinEncoding

Values — Ordinal gives one column per feature, holding its bin's index; scikit-learn's 'ordinal'. OneHot gives one column per bin of each feature, one of them set; 'onehot-dense', and the default shape.

Example — the same value, both ways.

using Lodestar.Preprocessing;

double[] ages = [19.0, 22.0, 25.0, 31.0, 38.0, 44.0, 52.0, 61.0, 67.0, 74.0];

KBinsDiscretizer hot = KBinsDiscretizer.Fit(
    ages, 1, new KBinsDiscretizerOptions { BinCount = 4 });

KBinsDiscretizer code = KBinsDiscretizer.Fit(
    ages, 1, new KBinsDiscretizerOptions { BinCount = 4, Encoding = BinEncoding.Ordinal });

string spread = string.Join(",", hot.Transform([44.0]));    // => 0,0,1,0
string index = string.Join(",", code.Transform([44.0]));    // => 2

Remarks — Ordinal states an order the data may not have. A linear model reading bin 3 as three times bin 1 is reading a claim the discretizer never made, which is why the reference defaults to one-hot and why this does too. Ordinal is right for a tree, which splits on the order and ignores the spacing, and for a column being written out for a human.

scikit-learn's third value, 'onehot', is absent: it returns a sparse matrix, and the difference is the storage rather than the encoding. A caller who wants the sparse form builds a CsrMatrix from the dense one.

Applies to — net10.0, netstandard2.0.

See also — KBinsDiscretizer.Transform, KBinsDiscretizerOptions, OneHotEncoder.