How to Load a class from file avoiding the set methods with empty arguments?

In the moment when I define the following test class, its constructor is called and with it the two set methods. However, in the moment when I save and then load the same class it seems that something more complicated is happening. By running the code below:

 Creating object
 Constructor
 ans =
     0
 ans =
     0
 Loading object from file
 ans =
     1
 ans =
     1
 ans =
     0
ans =
     0

Even if the constructor is not called (obviously, unless we use (ConstructOnLoad = true), two calls to the set methods are done to set the two properties to [].

Now, this is (from my perspective) a strange (and undesired) behavior because in set methods I would like to check that the argument passed is valid, but i don't want an empty array to be a valid argument. Ideally instead of having isempty(PAR) i want to have something like assert(PAR>0), but when doing so in the first call PAR will be [] and thus everything will crash.

Do you have any suggestion how to workaround this issue or in general if there is a nicer way to declare classes that will be saved and loaded from file?

I also tried using saveobj and loadobj, but it doesn't help.

Thank you -Roberto

-------------

classdef test < handle

    properties
        a = [];
        b = [];
    end

    methods

        function [OBJ] = test(arg1,arg2)
            fprintf('Constructor\n')
            switch nargin
                case {0}
                case {1}
                    if is(arg1,'test')
                        OBJ = arg1;
                    else
                        error('not a test class')
                    end
                otherwise
                    OBJ.a = arg1;
                    OBJ.b = arg2;
            end
        end

        function [] = set.a(OBJ,PAR)
            isempty(PAR)
            OBJ.a = PAR;
        end

        function [] = set.b(OBJ,PAR)
            isempty(PAR)
            OBJ.b = PAR;
        end
    end  
end

-------------

 fprintf('Creating object\n')
 A = test(1,2);
 save('file1')
 clearvars
 fprintf('Loading object from file\n')
 load('file1')

-------------

edit: it seems that the problem is indeed using

    properties
        a = [];
        b = [];
    end

to initialize the variables, for some reason the initialization at run-time don't make use of the set functions, but when loading, it does. I therefore changed the code to below and obtained an inverted behavior:

 Creating object
 Constructor
 ans =
     1
 ans =
     1
 ans =
     0
 ans =
     0
 Loading object from file
 ans =
     0
 ans =
     0

Hence, by initializing the properties with valid values I can hopefully get rid of this problem at loading time. Thanks for the help! :) ----------------

 classdef test < handle

    properties
        a
        b
    end

    methods

        function [OBJ] = test(arg1,arg2)
            fprintf('Constructor\n')

            switch nargin
                case {0}
                case {1}
                    if is(arg1,'test')
                        OBJ = arg1;
                    else
                        error('not a test class')
                    end
                otherwise
                    OBJ.default_values()
                    OBJ.a = arg1;
                    OBJ.b = arg2;

            end
        end

        function default_values(OBJ)
            OBJ.a = [];
            OBJ.b = [];
        end

        function [] = set.a(OBJ,PAR)
            isempty(PAR)
            OBJ.a = PAR;
        end

        function [] = set.b(OBJ,PAR)
            isempty(PAR)
            OBJ.b = PAR;
        end

    end

 end

[Matt J]

Now, this is (from my perspective) a strange (and undesired) behavior because in set methods I would like to check that the argument passed is valid, but i don't want an empty array to be a valid argument.

Although MATLAB doesn't force you to do this, I think it is far better to use property default values that your code will consider valid. So, if [] is not what you intend to treat as valid, you should be doing something like this,

    properties
        a = -inf;
        b = -inf;
    end

It is asking for trouble to allow an object to exist in any situation without valid property values.

[Matt J]

Note that your code has the line

>> clear vars

This will not clear A from memory. For that, the statement needs to be

>> clearvars %all one word

[Roberto Calandra]

You are right about the clearvars, fixed it in the code :)

Answer 1

[Oleg Komarov]

From tips from saving and loading:

"When loading an object, MATLAB® creates a new object and assigns the stored property values. For properties that had default values at the time you saved the object, MATLAB loads the saved default values, even if the class definition defines new default values for those properties."

So, simply remove the = [] from:

    a = []
    b = []

[Matt J]

I don't think this explains what we see in the OP's example. The saved property values were not the [] default values.

[Oleg Komarov]

It does. In his example, he calls the constructor, and that's it. Then saves and loads the object.

[Roberto Calandra]

I think that you are right Oleg, I updated the question with a possible solution. thanks :)

[Oleg Komarov]

You're right Matt, I misinterpreted the link. Also, I can't reproduce Roberto's problem on R2014a, which reminds me of a bug in oop with saving properties from a previous version of matlab.

Answer 2

[Matt J]

After some study, I think Oleg's answer is correct after all, but I still think the cited link makes this a bit murky. In particular, it doesn't describe how property set() methods behave during the load process. So, I'm providing an elaborated answer with illustrations using the classdef below,

    classdef myclass

     properties
       a=1;

     end

     methods

         function obj=myclass(val)

             obj.a=val;

         end

         function obj=set.a(obj,val) 
          testing = val,
          obj.a=val;
         end

     end

    end

Essentially, any explicit property defaults are saved to a .mat file along with the instance property values. When loaded back in, a skeleton object appears to be created first using the stored default property values. The stored defaults are then over-written by the stored instance values. In each of these steps, both the stored defaults and the stored instance values are passed through the current property set methods to ensure consistency with the current class definition. Hence,

    >> clear classes; A=myclass(2);save tst A;

    testing =

         2

    >> load tst

    testing =

         1

    testing =

         2

To verify this, I change the classdef below to have both a different default property value a=3 and also a different set.a() method.

    classdef myclass

     properties
       a=3;

     end

     methods

         function obj=myclass(val)

             obj.a=val;

         end

         function obj=set.a(obj,val) 
          testing = 10*val,
          obj.a=val;
         end

     end

    end

Note after "clear classes" and reloading that the new default value a=3 has no influence on the load behavior, but the new set.a() method does:

      >> clear classes; load tst

    testing =

        10

    testing =

        20

The slightly confusing thing (for me) is that when a default property value is not supplied explicitly in the classdef,

      classdef myclass

       properties
         a;

       end

       methods

           function obj=myclass

               constructorDefault = obj.a

           end

           function obj=set.a(obj,val) 
            testing = 10*val,
            obj.a=val;
           end

       end

      end

a default of [] is supplied by MATLAB during object construction, as clear from,

    >> clear classes; obj=myclass; obj.a=1; save tst obj

    constructorDefault =

         []

    testing =

        10

but a default of [] is not stored to the .mat file, as clear from

    >> load tst

    testing =

        10

Anyway, I guess this all makes sense since the load process isn't supposed to rely on the constructor...

[Oleg Komarov]

Very nice and clear. I think TMW should incorporate your examples (especially the order) in the documentation.

Answer 3

[James Tursa]

When you load a class object from a mat file, MATLAB must ensure that the object in the file matches the current definition of the class (i.e., properties match up). To do so it will call your class with no inputs, the intent being to get a skeleton of your class object back that it can use to compare properties with the mat file object. You need to be able to handle this behavior in your class constructor code. I don't know if this behavior can be changed or modified (I don't have the latest releases of MATLAB to test with).

[Matt J]

You need to be able to handle this behavior in your class constructor code.

I don't think so. That's what the ConstructOnLoad attribute is supposed to dictate.

Clearly, MATLAB does create a skeleton of the class using the initial property values in some way, but does not do so by calling the class constructor.

Also, it is not clear why this skeleton-creator calls the property set methods. Property default values are supposed to be exempt from the set method, as mentioned here.

[James Tursa]

Hmmm. I may be confusing the old style class with the classdef class, or maybe I just need a later version of MATLAB. (This used to be the behavior). Thanks for the correction.

Answer 4

[Xin Niu]

Hi, I have a more serious problem: when I load a object, a property is change to the default value even when I assigned a value to it before saving it. My code is like this:

classdef stim_info

      properties
          single_duration = [];% in seconds
          onset ＝ ［］；
          sample_rate = 10^4; % in Hz
          stim_thresh_ratio = 2/3; % used to determine onset sample.
          recording = []; 
          stim_status = []; 
          default_spike_thresh = 1; % parameters for function findspikes
          refractory_period = 4; % in milliseconds
      end
   methods

          % constructing methods:--------------------------------------------
          function SI = stim_info(varargin)
              if mod(nargin,2)
                  error('only pair wise input of name and value of parameters allowed')
              end

              SI.para_name=varargin(1:2:end);
              SI.para_value = cell2mat(varargin(2:2:end));
          end

          % set methods:-----------------------------------------------------
          function self = set.onset(self, onset)
              if self.check_onset(onset);
                  self.onset = onset(:);
              else
                  warning('onset not successfully defined')
              end
          end
  end

when I assign a value to onset and saved it, the value was empty when I loaded it. This problem happens when I did not set default value to that property.

[Matt J]

I think we need to know how to run the code, so as to reproduce what you're seeing.

[Xin Niu]

I write a brief code to reproduce the error:

if true
  % code
end
classdef test
  properties
      pro1 = [];
      pro2 = [];
  end

    methods
        function obj = test(p1,p2)
            if nargin>1; obj.pro1 = p1; end
            if nargin>2; obj.pro2 = p2; end
        end

        function self = set.pro1(self, p1)
            if self.check_prop(p1)
                self.pro1 = p1;
            end
        end
    end

    methods (Access=private)
        function isok = check_prop(self,p1)
            isok = length(p1)==length(self.pro2);
        end
    end
end

and the problem occurs when I run:

if true
  % code
end
t = test
t.pro2 = 4
t.pro1 = 3;
save t t
clear
load t
t.pro1
t.pro2