Development build. This page describes main, not a released package. The latest published Lodestar.Preprocessing is 0.2.0 — read its documentation.

Home › Preprocessing › Encoding and imputation

Which category OneHotEncoder drops, and what it does with an unseen value.

public sealed record OneHotEncoderOptions

Properties — Drop chooses which category loses its column, and Unknown what happens to a value the fit never saw — sklearn.preprocessing.OneHotEncoder's drop and handle_unknown, same defaults. MinFrequency makes a category seen fewer times infrequent, and MinFrequencyShare one seen in fewer than that share of the rows, strictly between 0 and 1: min_frequency as an integer and as a float, one setting, so the two are refused together. MaxCategories caps each feature's columns, its infrequent one included — max_categories. All three are null by default, which groups nothing.

Example — IfBinary drops the first category only where the feature has exactly two.

using Lodestar.Preprocessing;

var options = new OneHotEncoderOptions { Drop = CategoryDrop.IfBinary };

OneHotEncoder<string> binary = Encoders.OneHot(["y", "n", "y"], 1, options);
OneHotEncoder<string> three = Encoders.OneHot(["a", "b", "c"], 1, options);

int binaryColumns = binary.EncodedFeatureCount;  // => 1
int threeColumns = three.EncodedFeatureCount;    // => 3

Example — at most two columns: the most frequent category, and one for the rest.

using Lodestar.Preprocessing;

var capped = new OneHotEncoderOptions { MaxCategories = 2 };
OneHotEncoder<string> encoder = Encoders.OneHot(["a", "a", "a", "b", "b", "c", "d"], 1, capped);

int columns = encoder.EncodedFeatureCount;                            // => 2
string grouped = string.Join(",", encoder.InfrequentCategories[0]!);  // => b,c,d

Remarks — past MaxCategories the least frequent categories join the infrequent ones, ranked by a stable sort of the counts, so a tie keeps the sorted category order, as the reference's does. CategoryDrop.First drops the first column after grouping, and CategoryDrop.IfBinary counts the grouped columns. The drop and unknown settings interact: an ignored unknown encodes to all zeros, and so does a dropped first category, so a row of zeros means either. Transform says so with the example.

Applies to — net10.0, netstandard2.0.

See also — CategoryDrop, UnknownCategory.