mnrfit — Fit nominal multinomial logistic regression coefficients.

mnrfit(X,Y) fits nominal multinomial logistic regression coefficients for predictor matrix X and response Y. Rows of X are observations, columns are predictors, and columns of B correspond to all response classes except the final sorted reference class.

Syntax

B = mnrfit(X, Y)
B = mnrfit(X, Y, Name, Value)
[B, dev, stats] = mnrfit(X, Y)
[B, dev, stats] = mnrfit(X, Y, Name, Value)

Inputs

Returns

Returned values from mnrfit depend on how many outputs the caller requests.

Errors

How mnrfit works

• X must be a finite real numeric matrix with observations in rows and predictors in columns.
• Y may be a numeric, logical, string, character, cellstr, or categorical class vector with one response per row of X. Classes are sorted, and the final class is used as the reference category.
• Y may also be an N-by-K numeric grouped-counts matrix. Columns are treated as response categories, and row totals become multinomial observation weights.
• B = mnrfit(...) returns a coefficient matrix with one row per model term and one column per non-reference class. A constant term is included by default.
• [B,dev,stats] = mnrfit(...) returns model deviance and a stats structure containing beta, se, t, p, covb, coeffcorr, dfe, s, estdisp, and classNames.
• Model currently supports "nominal"; Link supports "logit"; Constant accepts logical values or "on"/"off"; Interactions accepts "off".
• Weights accepts a nonnegative finite observation-weight vector with positive total weight. IterationLimit and Tolerance control Newton solver convergence and may also be supplied through a statset-style Options struct. MaxIter and TolX are accepted as compatibility aliases.
• Rows with NaN in X, numeric Y, or grouped counts are omitted before fitting.
• Display is accepted for compatibility. Overdispersion estimation is not implemented, so EstDisp must be "off".

Examples

Fit a binary nominal logistic model

X = [0; 1; 2; 3; 4];
y = [1; 2; 1; 2; 2];
[B,dev,stats] = mnrfit(X, y)

Expected output:

B has two rows, one intercept and one predictor coefficient. stats.se has the same shape as B.

Fit a three-class nominal model

X = [0; 1; 0; 1; 2; 3];
y = [1; 1; 2; 2; 3; 3];
B = mnrfit(X, y, "Model", "nominal")

Expected output:

B has two columns because the final sorted class is the reference category.

Fit binomial count responses

X = [0; 1; 2; 3];
Y = [8 1 1; 5 3 2; 2 5 3; 1 6 4];
B = mnrfit(X, Y, "IterationLimit", 200)

Expected output:

Y columns contribute grouped response counts, and B has one column per non-reference category.

Fit text response labels

X = [0; 1; 2; 3; 4];
y = ["low"; "high"; "low"; "high"; "high"];
[B,dev,stats] = mnrfit(X, y, "Tolerance", 1e-7)

Expected output:

stats.classNames is returned as a string array containing the sorted response classes.

Using mnrfit with coding agents

Open a RunMat example with live inputs, then ask the agent to explain how mnrfit changes the result.

FAQ

Related Stats functions

Open-source implementation

Unlike proprietary runtimes, every RunMat function is open-source. Read exactly how mnrfit is executed, line by line, in Rust.

• View the source for mnrfit in Rust on GitHub
• Learn how the RunMat runtime works
• Found a bug? Open an issue with a minimal reproduction.

About RunMat

RunMat is an open-source runtime that executes MATLAB-syntax code blazing on any GPU. It is licensed under the Apache 2.0 license.

• RunMat automatically optimizes your math for GPU execution on Apple, Nvidia, and AMD hardware. No code changes needed. Simulations that took hours now take minutes.
• Start running code in seconds. RunMat runs in the browser, on the desktop, or from the CLI. No license server, no IT ticket.

Getting started · Benchmarks · Pricing