Deep learning: how to labels data using a datastore?

Hello,

I have some data (close to 500 *.mat files, size 3620x1, corresponding to measured signals) which are classified into two folders. Folder names represent the status of the data. I would like to perfom some deep learning on it: to do this, I tried to create a datastore using fileDatastore, but I don't know how to label all data with the folder name. Is it any possibility?

I know I can add 'LabelSource' using imageDatastore, but the Deep Network Designer did not work with such a datastore with *.mat files inside... Is anybody faced this issue?

Answer 1

[Ganapathi Subramanian R]

Hi Arnaud,

I understand that you are working on Deep Learning project and have an issue in labelling data using a datastore. Please provide the ‘FileExtensions’ and ‘ReadFcn’ Name-Value pairs in the ‘imageDatastore’ function to specify that the files are ‘.mat’ files and provide a custom function that can extract the matrix from each MAT-file.

dataPath = fullfile('data'); 
imds = imageDatastore(dataPath, ... 
    'IncludeSubfolders',true,'LabelSource','foldernames','FileExtensions',{'.mat'},'ReadFcn',@my_read_fcn); 

function res = my_read_fcn(f) 
s = load(f); 
res = s.NameOfMatrixVariable; 
end 

Please refer the below documentation for further information regarding ‘imageDatastore’ function.

https://www.mathworks.com/help/matlab/ref/matlab.io.datastore.imagedatastore.html