% Practicum, Task #3, 'Compositions of algorithms'.

%

% FUNCTION:

% [model] = bagging_train (X, y, num_iterations, base_algorithm, ...

%                   param_name1, param_value1, param_name2, param_value2)

%

% DESCRIPTION:

% This function train the composition of algorithms using bagging method.

%

% INPUT:

% X --- matrix of objects, N x K double matrix, N --- number of objects, 

%       K --- number of features.

% y --- vector of answers, N x 1 double vector, N --- number of objects. y

%       can have only two values --- +1 and -1.

% num_iterations --- the number ob algorithms in composition, scalar.

% base_algorithm --- the base algorithm, string. Can have one of two

%                    values: 'classification_tree' or 'svm'.

% param_name1 --- parameter of base_algorithm. For 'classification_tree' it 

%            is a 'min_parent' --- min number of objects in the leaf of 

%            classification tree. For 'svm' it is 'gamma' parameter.

% param_name2 --- parameter, that exists only for 'svm', it is a 'C' 

%                 parameter.

% param_value1, param_value2 --- values of corresponding parametres,

%                                scalar.

% OUTPUT:

% model --- trained composition, structure with two fields

%       - models --- cell array with trained models

%       - algorithm --- string, 'svm' or 'classification_tree'

%

% AUTHOR: 

% Murat Apishev (great-mel@yandex.ru)

%

function [model] = bagging_train (X, y, num_iterations, base_algorithm, ...

                        param_name1, param_value1, param_name2, param_value2)

    no_objects = size(X, 1);

    models = cell([1 num_iterations]);

    if strcmp(base_algorithm, 'svm')

        if ~strcmp(param_name1, 'gamma')

            temp = param_value1;

            param_value1 = param_value2;

            param_value2 = temp;

        end

        for iter = 1 : num_iterations

            indices = randi(no_objects, 1, no_objects);

            indices = unique(indices);

            models{iter} = svmtrain(y(indices), X(indices,:), ...

                [' -g ', num2str(param_value1), ' -c ', num2str(param_value2)]);

        end

    elseif strcmp(base_algorithm, 'classification_tree')

        for iter = 1 : num_iterations

            indices = randi(no_objects, 1, no_objects);

            indices = unique(indices);

            if (param_value1 > length(indices))

                value = length(indices);

            else

                value = param_value1;

            end

            models{iter} = ClassificationTree.fit(X(indices,:), y(indices), 'MinParent', value);

        end        

    else

        error('Incorrect type of algorithm!');

    end

    model.models = models;

    model.algorithm = base_algorithm;

end