limzunyuan

function[results] = ten_fold_cross_validation_parameter_estimation(x, y)
 
     results = [];
     
     % Train and pick the best tree using 10-fold cross validation.
     %[x2, y2] = ANNdata(x, y);
     for hidden_neurons_1 = 10:25
         hidden_neurons_1
         for hidden_neurons_2 = 10:25
             hidden_neurons_2

limzunyuan

function [ matrix ] = confusion_matrix(size, predicted, actual)
%CONFUSION_MATRIX Generates general confusion matrix
%   User inputs size of expected matrix, get length of vector input
%   For length of vector input, increment value of matrix with coordinates
%   corresponding to value in vector row.
    len = length(actual);
    matrix = zeros(size);
    predicted;
    for n = 1:len
        x = actual(n, :) + 1 ;

limzunyuan

% Accepts as input network (net), examples
% (x2, same format as x) and 
% combination method (combine_method) to 
% produce a vector of label predictions.
function [predictions] = testANN(net, x2, combine_method)
    predictions = sim(net, x2);
    %predictions = combine_method(out);
end

limzunyuan

% Accepts as input network (net), examples
% (x2, same format as x) and 
% combination method (combine_method) to 
% produce a vector of label predictions.
function [predictions] = testANN(net, x2, combine_method)
    predictions = sim(net, x2);
    %predictions = combine_method(out);
end

limzunyuan

https://gist.github.com/limzunyuan/c0791bd7c2eb40ec56bf

limzunyuan

function [ rate ] = avg_classification_rate( matrix )
%AVG_CLASSFICATION_RATE Calculates average classification rate
%   CR is no. of correctly classified examples (trace) divided by
%   total no. of examples (sum of matrix)
    total = sum(sum(matrix));
    correct = trace(matrix);
    rate = correct / total;
end

limzunyuan

function [result] = avg_precision_rate(matrix)
    dim = size(matrix, 1);
    result = zeros(1, dim);
    for n = 1:size(matrix, 1)
        col = matrix(:, n);
        tp = col(n);
        denom = sum(col);
        result(n) = tp/denom;
    end

limzunyuan

% Returns indices for training set and test set
% e.g. partition(1) => [1:900 , 901:1004]
function [training_idx testing_idx] = partition(ith_partition, x2)
    nrows = size(x2, 2);
    test_set_start_index = floor((ith_partition - 1) / 10 * nrows) + 1;
    test_set_end_index = floor(ith_partition / 10 * nrows);
    testing_idx = test_set_start_index:test_set_end_index;
    training_idx = setdiff(1:nrows, testing_idx);
end

limzunyuan

function [results] = f1_measure(matrix)
    precision = avg_precision_rate(matrix);
    recall = avg_recall_rate(matrix);
    
    dim = size(matrix, 1);
    results = zeros(1, dim);
    for i = 1 : dim
        results(i) = 2 * precision(i) * recall(i) / (precision(i) + recall(i));
    end        
end

limzunyuan

function [net] = create_traingd(x, y, list_num_hidden_neurons, ...
    num_epochs, train_idx, validation_idx, test_idx, lr)
    %[x2, y2] = ANNdata(x, y);
    net = feedforwardnet(list_num_hidden_neurons);
    net = configure(net, x, y);
	 net.trainFcn = 'traingd';
     net.divideFcn = 'divideind';
     net.divideParam.trainInd = train_idx;
     net.divideParam.valInd   = validation_idx;
     net.divideParam.testInd  = test_idx;