My MATLAB keeps reading old .mat files as uint32 even though it ain't, why is it and how can I fix this?

I thought of upgrading/updating my MATLAB, might that fix the issue?

Why is it happening in the first place?

As in:

load(data_set_path);

then if it had saved a struct there is doesn't return it but reads it as uint32 and throws error:

Warning: Variable 'data4cv' originally saved as a cross_validation_data
cannot be instantiated as an object and will be read in as a uint32. 
> In pca_train (line 5)

[James Tursa]

Looks like it was saved as an OOP object of type cross_validation_data, not a struct. Do you have the code for cross_validation_data?

[Brando Miranda]

Yes, I do:

https://github.com/brando90/research/blob/master/HBF_mat_lib/common/cross_validation/standard_train_cv_test_validation/cross_validation_data.m

or I can just copy it:

classdef cross_validation_data < handle
    %

      properties
          X
          y
          per_train
          per_cv
          X_train
          X_cv
          X_test
          y_train
          y_cv
          y_test
      end

      methods
          function obj = cross_validation_data(X,y,per_train,per_cv)
              obj.X = X;
              obj.y = y;
              obj.per_train = per_train;
              obj.per_cv = per_cv;
              %%split_data_for_hold_out_cross_validation
              [~, N] = size(obj.X);
              N_train = floor(N * obj.per_train);
              N_cv = floor(N * obj.per_cv);
              obj.X_train = obj.X(:,1:N_train);
              obj.X_cv = obj.X(:, N_train+1:N_train+N_cv );
              obj.X_test = obj.X(:, N_train+N_cv+1:N);
              obj.y_train = obj.y(:,1:N_train);
              obj.y_cv = obj.y(:, N_train+1:N_train+N_cv );
              obj.y_test = obj.y(:, N_train+N_cv+1:N);
          end
          function [ X_train,X_cv,X_test, y_train,y_cv,y_test ] = get_data_for_hold_out_cross_validation(obj)
              X_train = obj.X_train;
              X_cv = obj.X_cv;
              X_test = obj.X_test;
              y_train = obj.y_train;
              y_cv = obj.y_cv;
              y_test = obj.y_test;
          end
          function [] = change_data_sets(obj, X_train,X_cv,X_test, y_train,y_cv,y_test)
              obj.X_train = X_train;
              obj.X_cv = X_cv;
              obj.X_test = X_test;
              obj.y_train = y_train;
              obj.y_cv = y_cv;
              obj.y_test = y_test;
          end
          % Make a copy of a handle object.
          function new = copy(this)
              % Instantiate new object of the same class.
              new = feval(class(this));

              % Copy all non-hidden properties.
              p = properties(this);
              for i = 1:length(p)
                  new.(p{i}) = this.(p{i});
              end
          end
          function [] = normalize_data(obj)
              obj.X = normc(obj.X);
              obj.y = normc(obj.y);
              obj.X_train = normc(obj.X_train);
              obj.X_cv = normc(obj.X_cv);
              obj.X_test = normc(obj.X_test);
              obj.y_train = normc(obj.y_train);
              obj.y_cv = normc(obj.y_cv);
              obj.y_test = normc(obj.y_test);
          end
      end
      methods (Static)
          function [X_new, y_new] = shuffle_data(X,y)
              [~, N] = size(X);
              permute_ordering = randperm(N);
              X_new = X(:, permute_ordering);
              y_new = y(:,permute_ordering);
          end
      end

end

Answer 1

[Steven Lord]

You saved one or more objects in the MAT-file, but the definition for the object's class is not available to MATLAB when it loads the data so it doesn't know how to convert the raw data from the MAT-file into an instance of the object.

Try to remember if you were using any third-party toolboxes when you created that MAT-file. If so, check if they are still accessible to MATLAB. I'm betting that you were using such a toolbox but it is no longer available to MATLAB when you try to LOAD the file.

[Brando Miranda]

I was using my own code. In fact this is the defintion of the object:

https://github.com/brando90/research/blob/master/HBF_mat_lib/common/cross_validation/standard_train_cv_test_validation/cross_validation_data.m

classdef cross_validation_data < handle
    %

      properties
          X
          y
          per_train
          per_cv
          X_train
          X_cv
          X_test
          y_train
          y_cv
          y_test
      end

      methods
          function obj = cross_validation_data(X,y,per_train,per_cv)
              obj.X = X;
              obj.y = y;
              obj.per_train = per_train;
              obj.per_cv = per_cv;
              %%split_data_for_hold_out_cross_validation
              [~, N] = size(obj.X);
              N_train = floor(N * obj.per_train);
              N_cv = floor(N * obj.per_cv);
              obj.X_train = obj.X(:,1:N_train);
              obj.X_cv = obj.X(:, N_train+1:N_train+N_cv );
              obj.X_test = obj.X(:, N_train+N_cv+1:N);
              obj.y_train = obj.y(:,1:N_train);
              obj.y_cv = obj.y(:, N_train+1:N_train+N_cv );
              obj.y_test = obj.y(:, N_train+N_cv+1:N);
          end
          function [ X_train,X_cv,X_test, y_train,y_cv,y_test ] = get_data_for_hold_out_cross_validation(obj)
              X_train = obj.X_train;
              X_cv = obj.X_cv;
              X_test = obj.X_test;
              y_train = obj.y_train;
              y_cv = obj.y_cv;
              y_test = obj.y_test;
          end
          function [] = change_data_sets(obj, X_train,X_cv,X_test, y_train,y_cv,y_test)
              obj.X_train = X_train;
              obj.X_cv = X_cv;
              obj.X_test = X_test;
              obj.y_train = y_train;
              obj.y_cv = y_cv;
              obj.y_test = y_test;
          end
          % Make a copy of a handle object.
          function new = copy(this)
              % Instantiate new object of the same class.
              new = feval(class(this));

              % Copy all non-hidden properties.
              p = properties(this);
              for i = 1:length(p)
                  new.(p{i}) = this.(p{i});
              end
          end
          function [] = normalize_data(obj)
              obj.X = normc(obj.X);
              obj.y = normc(obj.y);
              obj.X_train = normc(obj.X_train);
              obj.X_cv = normc(obj.X_cv);
              obj.X_test = normc(obj.X_test);
              obj.y_train = normc(obj.y_train);
              obj.y_cv = normc(obj.y_cv);
              obj.y_test = normc(obj.y_test);
          end
      end
      methods (Static)
          function [X_new, y_new] = shuffle_data(X,y)
              [~, N] = size(X);
              permute_ordering = randperm(N);
              X_new = X(:, permute_ordering);
              y_new = y(:,permute_ordering);
          end
      end

end

[Walter Roberson]

Is that class in your MATLAB path at the time you try to do the load() ?

Answer 2

[Paul Kane]

If you wrote the app, open it in AppDesigner and run it. Back at the Matlab command window, you can now load your mat file and voila, Matlab now understands your structure.