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 › Encoding and imputation

SimpleImputerOptions

How SimpleImputer fills a missing value, and what to do with an empty feature.

public sealed record SimpleImputerOptions

Properties — Strategy chooses the statistic, FillValue is what ImputationStrategy.Constant fills with (zero by default), and KeepEmptyFeatures decides what happens to a feature with no value at all — sklearn.impute.SimpleImputer's strategy, fill_value and keep_empty_features.

Example — a constant the caller chose, and a feature that has nothing to average.

using Lodestar.Preprocessing;

double[] samples = [1.0, double.NaN, 3.0];

SimpleImputer constant = SimpleImputer.Fit(
    samples,
    1,
    new SimpleImputerOptions { Strategy = ImputationStrategy.Constant, FillValue = -1.0 });

string filled = string.Join(",", constant.Transform(samples));  // => 1,-1,3

Remarks — KeepEmptyFeatures is where this package and the reference differ. Its default drops an empty feature from the output; here that is refused, and this option fills the feature instead — the reference's keep_empty_features=True. It fills with zero, except under Constant, which fills with FillValue as the reference does. Either way the output has as many columns as the input, which the reference's default does not guarantee.

Applies to — net10.0, netstandard2.0.

See also — SimpleImputer.Fit, ImputationStrategy.