Unable to generate C code for Gaussian Process Regression Model

Hi Everyone

I am working on Battery State of Charge (SOC) Estimation using Gaussian Process Regression. I have successfully designed and optimized my model that works well according to my needs.

The Model Structure can be roughly described as:

Input: 6000x2 Array named (InputArray)

Column1 = capacity

Column 2 = voltage

Output: 6000 x 1 Array named (OutputArray)

Column1 = SOC

The GPR Model takes battery capacity and voltage as input and predicts the SOC as output. So far, the model provides high accuracy and I am able to use the predict function to observe and plot the responses.

My next task is to export the C code for this model, as I want to implement it on Texas Instrument CC1350 LaunchPad boards for my Wireless Sensor Network research. Here are the steps that I took to move towards the C/C++ Code Generation tool

• Saved the model using saveLearnerForCoder - Success

saveLearnerForCoder(trainedGPR.RegressionGP,'GPRout'

• Created The entry point function to load this model - Success

function label = entrypointGPR(X) %#codegen

Mdl = loadLearnerForCoder('GPRout');
label = predict(Mdl,X);
end

• codegen for MEX file - Error

testX = InputArray;
codegen entrypointGPR -args {testX} -report

This is where I am unable to figure out what to do next. I read the documentation on codegen, code builder and after many weeks I am still unable to figure out the next step. I am aware that perhaps I will not be able to use the code for my intended hardware (that's the worry for future project), but, I want to understand the complete code generation workflow. Ideally, if I am able to generate a desktop executable, I can still try to convert that C code for my intended hardware.

I would really appreciate if anyone could please point me in the right direction.

Bests

Answer 1

[Infinite_king]

Hi Omer Ali,

Here are a few basic things to know about the codegen command

"-args" option

Unlike MATLAB, C++ is a typed language, so the data type of the input should be determined before generating the code. For example, consider the following function

function res = add(x,y)
    res = x+y;
end

the function 'add' can be called with multiple data types.

add(double(2),double(3))
add(single(2),single(3))

However, while generating code the input data type should be fixed. To achieve this, the '-args' option should be used as follows,

% define the inputs
x = double(5);
y = double(6);

codegen function_name -args {x,y}
% note that the values in x and y doesn't matter, only their types are used
% to determine the input types

"-config" option

The MATLAB function can be converted to 'mex', static or dynamic library and an executable. To specify the resultant format and to customize few code generation setting "-config" option can be used as follows

cfg = coder.config('mex'); % for mex
cfg = coder.config('lib'); % for static library
cfg = coder.config('dll'); % for dynamic library

codegen function_name -args {arg1,arg2,arg3..} -config cfg

However, to generate an executable, a 'main' function is needed to serve as the entry point of the program. The config object can be configured to generate a sample main file,

cfg = coder.config('exe'); % for executable
cfg.GenerateExampleMain = true;

codegen function_name -args {arg1,arg2,arg3..} -config cfg

For more information please refer the following resources,

Generating Standalone C/C++ Executables from MATLAB Code - https://www.mathworks.com/help/coder/ug/standalone-c-c-executables-from-matlab-code.html

'codegen' command - https://www.mathworks.com/help/coder/ref/codegen.html

Hope this is helpful.