Main Content

predict

Predict responses of generalized linear regression model

Description

ypred = predict(mdl,Xnew) returns the predicted response values of the generalized linear regression model mdl to the points in Xnew.

example

[ypred,yci] = predict(mdl,Xnew) also returns confidence intervals for the responses at Xnew.

[ypred,yci] = predict(mdl,Xnew,Name,Value) specifies additional options using one or more name-value pair arguments. For example, you can specify the confidence level of the confidence interval.

example

Examples

collapse all

Create a generalized linear regression model, and predict its response to new data.

Generate sample data using Poisson random numbers with two underlying predictors X(:,1) and X(:,2).

rng('default') % For reproducibility
rndvars = randn(100,2);
X = [2 + rndvars(:,1),rndvars(:,2)];
mu = exp(1 + X*[1;2]);
y = poissrnd(mu);

Create a generalized linear regression model of Poisson data.

mdl = fitglm(X,y,'y ~ x1 + x2','Distribution','poisson');

Create data points for prediction.

[Xtest1,Xtest2] = meshgrid(-1:.5:3,-2:.5:2);
Xnew = [Xtest1(:),Xtest2(:)];

Predict responses at the data points.

ypred = predict(mdl,Xnew);

Plot the predictions.

surf(Xtest1,Xtest2,reshape(ypred,9,9))

Figure contains an axes object. The axes object contains an object of type surface.

Fit a generalized linear regression model, and then save the model by using saveLearnerForCoder. Define an entry-point function that loads the model by using loadLearnerForCoder and calls the predict function of the fitted model. Then use codegen (MATLAB Coder) to generate C/C++ code. Note that generating C/C++ code requires MATLAB® Coder™.

This example briefly explains the code generation workflow for the prediction of linear regression models at the command line. For more details, see Code Generation for Prediction of Machine Learning Model at Command Line. You can also generate code using the MATLAB Coder app. For details, see Code Generation for Prediction of Machine Learning Model Using MATLAB Coder App.

Train Model

Generate sample data using Poisson random numbers with two underlying predictors X(:,1) and X(:,2).

rng('default') % For reproducibility
rndvars = randn(100,2);
X = [2 + rndvars(:,1),rndvars(:,2)];
mu = exp(1 + X*[1;2]);
y = poissrnd(mu);

Create a generalized linear regression model. Specify the Poisson distribution for the response.

mdl = fitglm(X,y,'y ~ x1 + x2','Distribution','poisson');

Save Model

Save the fitted generalized linear regression model to the file GLMMdl.mat by using saveLearnerForCoder.

saveLearnerForCoder(mdl,'GLMMdl');

Define Entry-Point Function

In your current folder, define an entry-point function named mypredictGLM.m that does the following:

  • Accept new predictor input and valid name-value pair arguments.

  • Load the fitted generalized linear regression model in GLMMdl.mat by using loadLearnerForCoder.

  • Return predictions and confidence interval bounds.

function [yhat,ci] = mypredictGLM(x,varargin) %#codegen
%MYPREDICTGLM Predict responses using GLM model 
%   MYPREDICTGLM predicts responses for the n observations in the n-by-1
%   vector x using the GLM model stored in the MAT-file GLMMdl.mat,
%   and then returns the predictions in the n-by-1 vector yhat.
%   MYPREDICTGLM also returns confidence interval bounds for the
%   predictions in the n-by-2 vector ci.
CompactMdl = loadLearnerForCoder('GLMMdl');
narginchk(1,Inf);
[yhat,ci] = predict(CompactMdl,x,varargin{:});
end

Add the %#codegen compiler directive (or pragma) to the entry-point function after the function signature to indicate that you intend to generate code for the MATLAB algorithm. Adding this directive instructs the MATLAB Code Analyzer to help you diagnose and fix violations that would result in errors during code generation.

Generate Code

Generate code for the entry-point function using codegen (MATLAB Coder). Because C and C++ are statically typed languages, you must determine the properties of all variables in the entry-point function at compile time. To specify the data type and exact input array size, pass a MATLAB® expression that represents the set of values with a certain data type and array size. Use coder.Constant (MATLAB Coder) for the names of name-value pair arguments.

Create points for prediction.

[Xtest1,Xtest2] = meshgrid(-1:.5:3,-2:.5:2);
Xnew = [Xtest1(:),Xtest2(:)];

Generate code and specify returning 90% simultaneous confidence intervals on the predictions.

codegen mypredictGLM -args {Xnew,coder.Constant('Alpha'),0.1,coder.Constant('Simultaneous'),true}
Code generation successful.

codegen generates the MEX function mypredictGLM_mex with a platform-dependent extension.

If the number of observations is unknown at compile time, you can also specify the input as variable-size by using coder.typeof (MATLAB Coder). For details, see Specify Variable-Size Arguments for Code Generation and Specify Types of Entry-Point Function Inputs (MATLAB Coder).

Verify Generated Code

Compare predictions and confidence intervals using predict and mypredictGLM_mex. Specify name-value pair arguments in the same order as in the -args argument in the call to codegen.

[yhat1,ci1] = predict(mdl,Xnew,'Alpha',0.1,'Simultaneous',true);
[yhat2,ci2] = mypredictGLM_mex(Xnew,'Alpha',0.1,'Simultaneous',true);

The returned values from mypredictGLM_mex might include round-off differences compared to the values from predict. In this case, compare the values allowing a small tolerance.

find(abs(yhat1-yhat2) > 1e-6)
ans =

  0x1 empty double column vector
find(abs(ci1-ci2) > 1e-6)
ans =

  0x1 empty double column vector

The comparison confirms that the returned values are equal within the tolerance 1e–6.

Input Arguments

collapse all

Generalized linear regression model, specified as a GeneralizedLinearModel object created using fitglm or stepwiseglm, or a CompactGeneralizedLinearModel object created using compact.

New predictor input values, specified as a table, dataset array, or matrix. Each row of Xnew corresponds to one observation, and each column corresponds to one variable.

  • If Xnew is a table or dataset array, it must contain predictors that have the same predictor names as in the PredictorNames property of mdl.

  • If Xnew is a matrix, it must have the same number of variables (columns) in the same order as the predictor input used to create mdl. Note that Xnew must also contain any predictor variables that are not used as predictors in the fitted model. Also, all variables used in creating mdl must be numeric. To treat numerical predictors as categorical, identify the predictors using the 'CategoricalVars' name-value pair argument when you create mdl.

Data Types: single | double | table

Name-Value Arguments

Specify optional pairs of arguments as Name1=Value1,...,NameN=ValueN, where Name is the argument name and Value is the corresponding value. Name-value arguments must appear after other arguments, but the order of the pairs does not matter.

Before R2021a, use commas to separate each name and value, and enclose Name in quotes.

Example: [ypred,yci] = predict(Mdl,Xnew,'Alpha',0.01,'Simultaneous',true) returns the confidence interval yci with a 99% confidence level, computed simultaneously for all predictor values.

Significance level for the confidence interval, specified as the comma-separated pair consisting of 'Alpha' and a numeric value in the range [0,1]. The confidence level of yci is equal to 100(1 – Alpha)%. Alpha is the probability that the confidence interval does not contain the true value.

Example: 'Alpha',0.01

Data Types: single | double

Number of trials for the binomial distribution, specified as the comma-separated pair consisting of 'BinomialSize' and a scalar or vector of the same length as the response. predict expands the scalar input into a constant array of the same size as the response. The scalar input means that all observations have the same number of trials.

The meaning of the output values in ypred depends on the value of 'BinomialSize'.

  • If 'BinomialSize' is 1 (default), then each value in the output ypred is the probability of success.

  • If 'BinomialSize' is not 1, then each value in the output ypred is the predicted number of successes in the trials.

Data Types: single | double

Offset value for each row in Xnew, specified as the comma-separated pair consisting of 'Offset' and a scalar or vector with the same length as the response. predict expands the scalar input into a constant array of the same size as the response.

Note that the default value of this argument is a vector of zeros even if you specify the 'Offset' name-value pair argument when fitting a model. If you specify 'Offset' for fitting, the software treats the offset as an additional predictor with a coefficient value fixed at 1. In other words, the formula for fitting is

f(μ) = Offset + X*b,

where f is the link function, μ is the mean response, and X*b is the linear combination of predictors X. The Offset predictor has coefficient 1.

Data Types: single | double

Flag to compute simultaneous confidence bounds, specified as the comma-separated pair consisting of 'Simultaneous' and either true or false.

  • truepredict computes confidence bounds for the curve of response values corresponding to all predictor values in Xnew, using Scheffé's method. The range between the upper and lower bounds contains the curve consisting of true response values with 100(1 – α)% confidence.

  • falsepredict computes confidence bounds for the response value at each observation in Xnew. The confidence interval for a response value at a specific predictor value contains the true response value with 100(1 – α)% confidence.

With simultaneous bounds, the entire curve of true response values is within the bounds at high confidence. By contrast, non-simultaneous bounds require only the response value at a single predictor value to be within the bounds at high confidence. Therefore, simultaneous bounds are wider than non-simultaneous bounds.

Example: 'Simultaneous',true

Output Arguments

collapse all

Predicted response values at Xnew, returned as a numeric vector.

For a binomial model, the meaning of the output values in ypred depends on the value of the 'BinomialSize' name-value pair argument.

  • If 'BinomialSize' is 1 (default), then each value in the output ypred is the probability of success.

  • If 'BinomialSize' is not 1, then each value in the output ypred is the predicted number of successes in the trials.

For a model with an offset, specify the offset value by using the 'Offset' name-value pair argument. Otherwise, predict uses 0 as the offset value.

Confidence intervals for the responses, returned as a two-column matrix with each row providing one interval. The meaning of the confidence interval depends on the settings of the name-value pair arguments 'Alpha' and 'Simultaneous'.

Alternative Functionality

  • feval returns the same predictions as predict. The feval function does not support the 'Offset' and 'BinomialSize' name-value pair arguments. feval uses 0 as the offset value, and the output values in ypred are predicted probabilities. The feval function can take multiple input arguments for new predictor input values, with one input for each predictor variable, which is simpler to use with a model created from a table or dataset array. Note that the feval function does not give confidence intervals on its predictions.

  • random returns predictions with added noise.

Extended Capabilities

Version History

Introduced in R2012a