Home › Stats-Regression › Multinomial logit

MultinomialLogitOptions

What a multinomial logit fit should estimate, at what confidence, and how long Newton may run.

public sealed record MultinomialLogitOptions

Properties — WithIntercept prepends a column of ones to every equation; true by default. ConfidenceLevel is the two-sided level of the intervals; 0.95 by default, strictly inside (0, 1). MaximumIterations is Newton's budget; 35 by default, the reference's. Tolerance is the largest step in any parameter that still counts as converged; 1e-8 by default. ThrowOnNonConvergence says whether a fit that spends its budget throws; true by default.

Example — a fit allowed one step, inspected rather than refused.

using Lodestar.Stats.Regression;

double[] design = [-1.1, -0.73, -0.78, 0.27, -0.25, 0.13, 0.84, 0.86, 0.48, -0.45, -0.75, -0.81, -0.34, -0.05, -0.97];
int[] response = [2, 1, 0, 0, 1, 0, 2, 2, 0, 1, 1, 2, 1, 0, 1];

MultinomialLogitSummary once = MultinomialLogit.Fit(
    design, response, 1, new MultinomialLogitOptions { MaximumIterations = 1, ThrowOnNonConvergence = false });

bool converged = once.Converged;  // => False

Remarks — converged is false exactly when the budget was spent, as in the reference: a fit whose last allowed step was already small enough still reports false.

Applies to — net10.0, netstandard2.0.

See also — MultinomialLogit.Fit, MultinomialLogitSummary.