sigmf doesnt' work in octave ?

hello l get this error when l execute this code :

error: 'sigmf' undefined near line 29 column 28

and sigmf is predefined in octave and matlab as it is mentionned in octave and matlab documentation

        % load training set and testing set
        clear all;
        train_set = loadMNISTImages('/home/anelmad/my_codes/MNIST_digit_recognition-master/load_data/train-images.idx3-ubyte');
        train_label = loadMNISTLabels('/home/anelmad/my_codes/MNIST_digit_recognition-master/load_data/train-labels.idx1-ubyte');
        test_set = loadMNISTImages('/home/anelmad/my_codes/MNIST_digit_recognition-master/load_data/t10k-images.idx3-ubyte');
        test_label = loadMNISTLabels('/home/anelmad/my_codes/MNIST_digit_recognition-master/load_data/t10k-labels.idx1-ubyte');

        % parameter setting
        alpha = 0.1; % learning rate
        beta = 0.01; % scaling factor for sigmoid function
        train_size = size(train_set);
        N = train_size(1); % number of training samples
        D = train_size(2); % dimension of feature vector
        n_hidden = 300; % number of hidden layer units
        K = 10; % number of output layer units
        % initialize all weights between -1 and 1
        W1 = 2*rand(1+D, n_hidden)-1; % weight matrix from input layer to hidden layer
        W2 = 2*rand(1+n_hidden, K)-1; % weight matrix from hidden layer to ouput layer
        max_iter = 100; % number of iterations
        Y = eye(K); % output vector 

        % training 
        for i=1:max_iter
            disp([num2str(i), ' iteration']);
            for j=1:N
                % propagate the input forward through the network
                input_x = [1; train_set(j, :)'];
                hidden_output = [1;sigmf(W1'*input_x, [beta 0])];
                output = sigmf(W2'*hidden_output, [beta 0]);
                % propagate the error backward through the network
                % compute the error of output unit c
                delta_c = (output-Y(:,train_label(j)+1)).*output.*(1-output);
                % compute the error of hidden unit h
                delta_h = (W2*delta_c).*(hidden_output).*(1-hidden_output);
                delta_h = delta_h(2:end);
                % update weight matrix
                W1 = W1 - alpha*(input_x*delta_h');
                W2 = W2 - alpha*(hidden_output*delta_c');
            end
        end

        % testing 
        test_size = size(test_set);
        num_correct = 0;
        for i=1:test_size(1)
            input_x = [1; test_set(i,:)'];
            hidden_output = [1; sigmf(W1'*input_x, [beta 0])];
            output = sigmf(W2'*hidden_output, [beta 0]);
            [max_unit, max_idx] = max(output);
            if(max_idx == test_label(i)+1)
                num_correct = num_correct + 1;
            end
        end
        % computing accuracy
        accuracy = num_correct/test_size(1);

or l have to define sigmf ?

Thanks for helps

[Stephen23]

The sigmf function requires the Fuzzy Logic Toolbox: do you have this installed?

[MiMad]

l haven't installed it. do you mind telling me how l can install it? since l'm new to octave, l'll be so grateful

[Stephen23]

Are you using Octave or MATLAB?

[MiMad]

l'm using octave

Answer 1

[Walter Roberson]

sigmf is part of the MATLAB Fuzzy Logic Toolbox; http://www.mathworks.com/help/fuzzy/sigmf.html

If you do not have that toolbox you might be able to substitute http://www.mathworks.com/matlabcentral/fileexchange/47204-fuzzy-model-identification-for-control/content/E5_2_IMC_Liq/sigmf.m

[MiMad]

how can l install Fuzzy Logic Toolbox in octave ?

[Walter Roberson]

To install Fuzzy Logic Toolbox, you upgrade Octave to MATLAB.

Note: this is a MATLAB forum, so Octave installation or capability questions are off topic and should be asked in an Octave forum.

[MiMad]

l installed successfully Fuzzy Logic Toolbox on octave. l tried a trivial example to test sigmf whether it works :

 >>> x= 0:100; >>> params=[0.3 40]; y1= sigmf(x,params);

And definitely it works. But when l try to execute the code above it returns me this error :

 error: sigmf's first argument must be a valid domain

corresponding to this line of code

   output = sigmf(W2'*hidden_output, [beta 0]);

How to circumvent this error ?

Thank you

[Walter Roberson]

The Mathworks Fuzzy Logic Toolbox cannot be used with Octave, not within the terms of the license agreement.

There is an open source Octave Fuzzy Logic Toolkit. If that is what you are using, you should be asking the author or asking in an Octave resource; this forum is a MATLAB resource and Octave is off-topic here.

[Franklin Yu]

The Octave package is also available in Octave-Forge.