predict class labels using matlab function block -凯发k8网页登录

train classification model

this example uses the ionosphere data set, which contains radar-return qualities (y) and predictor data (x). radar returns are either of good quality ('g') or of bad quality ('b').

load the ionosphere data set. determine the sample size.

load ionosphere
n = numel(y)

n = 351

the matlab function block cannot return cell arrays. convert the response variable to a logical vector whose elements are 1 if the radar returns are good, and 0 otherwise.

y = strcmp(y,'g');

suppose that the radar returns are detected in sequence, and you have the first 300 observations, but you have not received the last 51 yet. partition the data into present and future samples.

prsntx = x(1:300,:);
prsnty = y(1:300);
ftrx = x(301:end,:);
ftry = y(301:end);

train an svm model using all, presently available data. specify predictor data standardization.

mdl = fitcsvm(prsntx,prsnty,'standardize',true);

mdl is a classificationsvm object, which is a linear svm model. the predictor coefficients in a linear svm model provide enough information to predict labels for new observations. removing the support vectors reduces memory usage in the generated code. remove the support vectors from the linear svm model by using the function.

mdl = discardsupportvectors(mdl);

save model using savelearnerforcoder

at the command line, you can use mdl to make predictions for new observations. however, you cannot use mdl as an input argument in a function meant for code generation.

prepare mdl to be loaded within the function using savelearnerforcoder.

savelearnerforcoder(mdl,'svmionosphere');

savelearnerforcoder compacts mdl, and then saves it in the mat-file svmionosphere.mat.

define matlab function

define a matlab function named svmionospherepredict.m that predicts whether a radar return is of good quality. the function must:

function [label,score] = svmionospherepredict(x) %#codegen
%svmionospherepredict predict radar-return quality using svm model
%   svmionospherepredict predicts labels and estimates classification
%   scores of the radar returns in the numeric matrix of predictor data x
%   using the compact svm model in the file svmionosphere.mat.  rows of x
%   correspond to observations and columns to predictor variables.  label
%   is the predicted label and score is the confidence measure for
%   classifying the radar-return quality as good.
%
% 凯发官网入口首页 copyright 2016 the mathworks inc.
mdl = loadlearnerforcoder('svmionosphere');
[label,bothscores] = predict(mdl,x);
score = bothscores(:,2);
end

note: if you click the button located in the upper-right section of this page and open this example in matlab, then matlab opens the example folder. this folder includes the entry-point function file.

create simulink model

create a simulink model with the matlab function block that dispatches to svmionospherepredict.m.

this example provides the simulink model slexsvmionospherepredictexample.slx. open the simulink model.

simmdlname = 'slexsvmionospherepredictexample'; 
open_system(simmdlname)

the figure displays the simulink model. when the input node detects a radar return, it directs that observation into the matlab function block that dispatches to svmionospherepredict.m. after predicting the label and score, the model returns these values to the workspace and displays the values within the model one at a time. when you load slexsvmionospherepredictexample.slx, matlab also loads the data set that it requires called radarreturninput. however, this example shows how to construct the required data set.

the model expects to receive input data as a structure array called radarreturninput containing these fields:

create an appropriate structure array for future radar returns.

radarreturninput.time = (0:50)';
radarreturninput.signals(1).values = ftrx;
radarreturninput.signals(1).dimensions = size(ftrx,2);

you can change the name from radarreturninput, and then specify the new name in the model. however, simulink expects the structure array to contain the described field names.

simulate the model using the data held out of training, that is, the data in radarreturninput.

sim(simmdlname);

the figure shows the model after it processes all observations in radarreturninput one at a time. the predicted label of x(351,:) is 1 and its positive-class score is 1.431. the variables tout, yout, and svmlogsout appear in the workspace. yout and svmlogsout are simulinkdata.dataset objects containing the predicted labels and scores. for more details, see data format for logged simulation data (simulink).

extract the simulation data from the simulation log.

labelssl = svmlogsout.getelement(1).values.data;
scoressl = svmlogsout.getelement(2).values.data;

labelssl is a 51-by-1 numeric vector of predicted labels. labelssl(j) = 1 means that the svm model predicts that radar return j in the future sample is of good quality, and 0 means otherwise. scoressl is a 51-by-1 numeric vector of positive-class scores, that is, signed distances from the decision boundary. positive scores correspond to predicted labels of 1, and negative scores correspond to predicted labels of 0.

predict labels and positive-class scores at the command line using predict.

[labelcmd,scorescmd] = predict(mdl,ftrx);
scorescmd = scorescmd(:,2);

labelcmd and scorescmd are commensurate with labelssl and scoressl.

compare the future-sample, positive-class scores returned by slexsvmionospherepredictexample to those returned by calling predict at the command line.

err = sum((scorescmd - scoressl).^2);
err < eps

ans = logical
   1

the sum of squared deviations between the sets of scores is negligible.

if you also have a simulink coder™ license, then you can generate c code from slexsvmionospherepredictexample.slx in simulink or from the command line using slbuild (simulink). for more details, see generate c code for a model (simulink coder).