Bug in loadLearnerForCoder from MATLAB coder

2 visualizaciones (últimos 30 días)
Diogo Tecelão
Diogo Tecelão el 1 de Dic. de 2020
Respondida: Marou AD el 11 de Feb. de 2024
Hello,
I've built a classification model (esemble), and saved it using saveLearnerForCoder. However, when compiling my code to C using MATLAB coder, an error is raised. I'm using MATLAB 2020a. This issue appears to have been previously reported here. Please, find detailed information below. I would very appreciate a workaround.
I'm loading the classifier as recommended:
cough_detection_model = loadLearnerForCoder('my_model'); % I tried with both ' and ", just in case
This triggers the following error:
??? The input to coder.const cannot be reduced to a constant: Unsupported value.
Error in ==> loadLearnerForCoder Line: 41 Column: 68
My classifier was created with the following code:
template_tree = templateTree('MaxNumSplits', 700, 'MinLeafSize', 20);
ensemble_model = fitcensemble(features, classes, ...
'Learners', template_tree, 'Method','RUSBoost', 'NumLearningCycles', 500, ...
'LearnRate', 0.1, 'ScoreTransform', 'None');
saveLearnerForCoder(ensemble_model, 'my_model')
I am able to load the classifier and use it normally. The error only happens when using the codegen.
I made some digging, and created a bypass function which avoids the problems from loadLearnerForCoder. Obviously it would crash somewhere, but I think it gave me some further information. This is the code I managed to create:
matFile = coder.load("my_model");
fromStructFcn = str2func('classreg.learning.classif.CompactClassificationEnsemble.fromStruct');
obj = fromStructFcn(matFile.compactStruct); % raises error
Which raises the following error:
??? Failed to compute constant value for nontunable property 'NumPredictors'. In code generation, nontunable
properties can only be assigned constant values.
I have tried to overpass this by doing:
matFile.compactStruct.DataSummary.NumPredictors = coder.const(117);
But this doesn't help either.
Can please anyone help me?
  1 comentario
Diogo Tecelão
Diogo Tecelão el 1 de Dic. de 2020
Update
By using the following code:
fromStructFcn = str2func('classreg.learning.classif.CompactClassificationEnsemble.fromStruct');
obj = fromStructFcn(coder.const(matFile.compactStruct));
I now get a different error:
??? Code generation does not support MATLAB class 'classreg.learning.coder.classif.CompactClassificationEnsemble'
in this context because the class does not implement toRedirect() and/or fromRedirect() methods.

Iniciar sesión para comentar.

Respuestas (2)

Sriram Padavettan
Sriram Padavettan el 4 de Dic. de 2020
Editada: Sriram Padavettan el 4 de Dic. de 2020
The problem is 2020A. I think there's a bug in that version. I messed around with this same issue all day yesterday on 2020A, tried 2020B instead, and it worked. Maybe even 2019b or earlier would work.
:)

Marou AD
Marou AD el 11 de Feb. de 2024
It seems there is a bug in MATLAB 2020a that might be causing the problem. I had a similar issue, and after experimenting, I found that switching to MATLAB 2020b resolved the problem for me. It seems there might be a compatibility or bug fix in the later version. I recommend giving MATLAB 2020b a try , it will resolves your compilation issue.

Productos


Versión

R2020a

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by