bag_mi_svm.m

包含了很多分类算法

function [test_bag_label, test_inst_label, test_bag_prob, test_inst_prob] = bag_MI_SVM(para, train_bags, test_bags)

global preprocess;
global temp_train_file temp_test_file temp_output_file temp_model_file libSVM_dir; 

num_train_bag = length(train_bags);
num_test_bag = length(test_bags);

[train_instance, dummy] = bag2instance(train_bags);
[test_instance, dummy] = bag2instance(test_bags);

if isempty(train_bags)

    if (~isfield(preprocess, 'model_file') || isempty(preprocess.model_file))
        error('The model file must be provided in the train_only setting!');
    end;
    eval(['!copy ' preprocess.model_file ' ' temp_model_file ]);
    [test_label_predict, test_prob_predict] = LibSVM(para, [], [], test_instance, ones(num_test_inst, 1));    
else
    %set the initial instance labels to bag labels
    idx = 0;
    num_pos_train_bag = 0;
    for i = 1: num_train_bag
        num_inst = size(train_bags(i).instance, 1);
        if train_bags(i).label == 0
            sample_instance(idx+1 : idx+num_inst, :) = train_bags(i).instance;
            sample_label(idx+1 : idx+num_inst) = zeros(num_inst, 1);
            idx = idx + num_inst;
        else
            sample_instance(idx+1, :) = mean(train_bags(i).instance, 1);
            sample_label(idx+1) = 1;
            idx = idx + 1;
            num_pos_train_bag = num_pos_train_bag + 1;
        end
    end

    num_train_inst = size(train_instance, 1);
    num_test_inst = size(test_instance, 1);

    num_neg_train_bag = num_train_bag - num_pos_train_bag;
    num_neg_train_inst = idx - num_pos_train_bag;
    avg_num_inst = num_neg_train_inst / num_neg_train_bag;

    selection = zeros(num_pos_train_bag, 1);
    step = 1;
    past_selection(:, 1) = selection;    
    
    while 1
        new_para = sprintf(' -NegativeWeight %.10g', 1/avg_num_inst);
        [all_label_predict, all_prob_predict] = LibSVM([para new_para], sample_instance, sample_label, [train_instance; test_instance], ones(num_train_inst+num_test_inst, 1));
        train_label_predict = all_label_predict(1 : num_train_inst);
        train_prob_predict = all_prob_predict(1 : num_train_inst);
        test_label_predict = all_label_predict(num_train_inst+1 : num_train_inst+ num_test_inst);
        test_prob_predict = all_prob_predict(num_train_inst+1 : num_train_inst+ num_test_inst);

        idx = 0;
        pos_idx = 1;
        for i=1:num_train_bag
            num_inst = size(train_bags(i).instance, 1);

            if train_bags(i).label == 0
                sample_instance(idx+1 : idx+num_inst, :) = train_bags(i).instance;
                sample_label(idx+1 : idx+num_inst) = zeros(num_inst, 1);                
                idx = idx + num_inst;
            else
                [sort_prob, sort_idx] = sort(train_prob_predict(idx+1 : idx+num_inst));
                sample_instance(idx + 1, :) = train_bags(i).instance(sort_idx(num_inst),:);
                sample_label(idx + 1) = 1;
                
                selection(pos_idx) = sort_idx(num_inst);
                pos_idx = pos_idx + 1;
                
                idx = idx + 1;
            end
        end

        %compare the current selection with previous selection, quit if same
        difference = sum(past_selection(:, step) ~= selection);
        fprintf('Number of selection changes is %d\n', difference);
        if difference == 0, break; end;
        
        repeat_selection = 0;
        for i = 1 : step
            if all(selection == past_selection(:,i))
                repeat_selection = 1;
                break;
            end               
        end

        if repeat_selection == 1
            fprintf('Repeated training selections found, quit...\n');
            break; 
        end

        step = step + 1;
        past_selection(:, step) = selection;
    end
end

test_inst_label = test_label_predict;
test_inst_prob = test_prob_predict;

idx = 0;
for i=1:num_test_bag
    num_inst = size(test_bags(i).instance, 1);
    test_bag_label(i) = any(test_inst_label(idx+1 : idx+num_inst));
    test_bag_prob(i) = max(test_inst_prob(idx+1 : idx+num_inst));
    idx = idx + num_inst;
end