Preprocessing binstrategy

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

Home › Preprocessing › Feature transforming

BinStrategy

Where KBinsDiscretizer places its bin edges.

public enum BinStrategy

Values — Uniform cuts equal widths between the feature's smallest and largest value; scikit-learn's 'uniform'. Quantile cuts equal counts, read off the feature's percentiles; 'quantile', and the default. KMeans cuts midway between one-dimensional k-means centres; 'kmeans'.

Example — the same ten ages, three 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];

var options = new KBinsDiscretizerOptions { BinCount = 4, Encoding = BinEncoding.Ordinal };

KBinsDiscretizer byCount = KBinsDiscretizer.Fit(ages, 1, options);
KBinsDiscretizer byWidth = KBinsDiscretizer.Fit(
    ages, 1, options with { Strategy = BinStrategy.Uniform });

string equalCounts = string.Join(",", byCount.BinEdges[0]);   // => 19,25,41,61,74

double firstWidthEdge = byWidth.BinEdges[0][1];               // => 32.75
double secondWidthEdge = byWidth.BinEdges[0][2];              // => 46.5

Remarks — which one depends on what the bin is for. Quantile guarantees every bin is equally populated, which is what a model wants and what makes the bins comparable across features; Uniform guarantees every bin covers the same range, which is what a reader wants when the bin will be printed as "30 to 45". On a skewed column the two disagree sharply: here the first uniform bin holds four of the ten ages and the third holds one.

KMeans is the only strategy that costs an iteration. It runs Lloyd's algorithm over the single column — Lodestar.Cluster's, which is the package edge this member added — starting from the uniform bin midpoints, then puts each edge midway between two consecutive centres. Starting from the midpoints rather than a random draw is what makes it deterministic, and the reference does the same.

Applies to — net10.0, netstandard2.0.

See also — KBinsDiscretizerOptions, KBinsDiscretizer.Fit.