How to read a 'CompactSVMModel' saved in a .mat file from a C++/Qt program

I tried to save a machine learning model into a .mat file, and then read it from a Qt/C++ application. Following the documentation, I managed to read files containing basic types like double or int. As my model is of type 'classreg.learning.classif.CompactClassificationSVM', I wonder if it is still possible to read it.

As and example, i saved this iosnosphere model on matlab

load ionosphere
SVMModel = fitcsvm(X,Y,'Standardize',true,'ClassNames',{'b','g'})
CompactSVMModel = compact(SVMModel)
save("CompactSVMModel.mat", "CompactSVMModel")

I imported the relevant library in the CMakeLists.txt

find_package(Matlab "x.xx" EXACT COMPONENTS  "MX_LIBRARY" "MAT_LIBRARY" "DATAARRAY_LIBRARY" )
target_link_libraries(target PRIVATE Matlab::mat )

I used the "mat.h" library to open and read the data.

#include "mat.h"

int main () {
std::string stdFilename = "/filepath/CompactSVMModel.mat";
const char *variableName = "CompactSVMModel";

// Open the MAT-File
MATFile *matFile = matOpen(stdFilename.c_str(), "r");
if (!matFile) {
    qDebug() << "Error opening MAT file: " << stdFilename;
}
    
// Read the variable from the MAT-File
mxArray *dataArray = matGetVariable(matFile, variableName);
if (!dataArray) {
    qDebug() << "Error reading variable: " << variableName << " from MAT file: " << stdFilename;
}

...
// read a save the diffrent fields into containers
           
return 0; 
}

At the first if statement, matFile is NULL.

Answer 1

[Arnav]

Hi,

I tried executing your code with the particulars provided in the CMakeLists.txt with MATLAB R2022b libraries.

This is my CMakeLists.txt:

cmake_minimum_required(VERSION 3.10)
project(ML_Answers)
find_package(Matlab "9.13.0.2553342"EXACT COMPONENTS MX_LIBRARY MAT_LIBRARY DATAARRAY_LIBRARY)
include_directories(${Matlab_INCLUDE_DIRS})
add_executable(ML_Answers main.cpp)
target_link_libraries(ML_Answers PRIVATE Matlab::mat)

I built and ran the project using the following commands:

$ cmake -G "MinGW Makefiles" -DCMAKE_CXX_COMPILER=C:\ProgramData\MATLAB\SupportPackages\R2022b\3P.instrset\mingw_w64.instrset\bin\g++.exe .
$ mingw32-make.exe
$./ML_Answers.exe

This executes as expected. I encountered the issue you mentioned when I built the project in a different build subdirectory (instead of the project root directory) and tried executing the executable from inside the build subdirectory. This was due to the relative path of the CompactSVMModel.mat file not being correct. You might try rechecking to see if you made a similar error.

For the task of reading a class instance from CompactSVMModel.mat file using a standalone C++ program, a workaround would be to save the different properties as different variables in the CompactSVMModel.mat file as follows:

% Extract the SupportVectors property
SupportVectors = CompactSVMModel.SupportVectors;
save('CompactSVMModel.mat', 'SupportVectors', 'CompactSVMModel');

Following this, the SupportedVectors variable can be read using C++ program in the following way (I substituted qDebug() for std::cerr and std::cout):

#include "mat.h"
#include <string>
#include <iostream>

int main() {
    std::string stdFilename = "./CompactSVMModel.mat";
    const char* variableName = "SupportVectors"; // Change to read the SupportVectors property
    
    // Open the MAT-File
    MATFile *matFile = matOpen(stdFilename.c_str(), "r");
    if (!matFile) {
        std::cerr << "Error opening MAT file: " << stdFilename << std::endl;
        return 1; // Exit with error code
    }
    
    // Read the variable from the MAT-File
    mxArray *dataArray = matGetVariable(matFile, variableName);
    if (!dataArray) {
        std::cerr << "Error reading variable: " << variableName << " from MAT file: " << stdFilename << std::endl;
        matClose(matFile);
        return 1; // Exit with error code
    }
        
    // Print the dimensions of the SupportVectors matrix
    mwSize numRows = mxGetM(dataArray);
    mwSize numCols = mxGetN(dataArray);
    std::cout << "SupportVectors dimensions: " << numRows << " x " << numCols << std::endl;
        
    // Clean up and close the MAT-File
    mxDestroyArray(dataArray);
    matClose(matFile);
        
    std::cout << "SupportVectors have been successfully read from the MAT file." << std::endl;
        
    return 0; 
}

Output for the C++ Program:

SupportVectors dimensions: 89 x 2

SupportVectors have been successfully read from the MAT file.

You can learn more about how to read variables here:

https://www.mathworks.com/help/releases/R2022b/matlab/matlab-c-api-to-read-mat-file-data.html?s_tid=CRUX_lftnav

Alternatively, you can build your project using MEX and use C++ MEX API. You can read more about the C++ MEX API here:

https://www.mathworks.com/help/releases/R2022b/matlab/matlab_external/cpp-mex-api.html#responsive_offcanvas

[MahnnUser]

Hi @Arnav,

Thank you for your response. In the end, I used a different approach. My goal was to translate my MATLAB program into a C++ program using MATLAB Coder. Since the MATLAB program used a model, I needed to import it. However, I found that the model could be integrated into the C++ library generated by MATLAB Coder. https://fr.mathworks.com/help/stats/loadlearnerforcoder.html