ConditionalProbabilityTree.cpp

Go to the documentation of this file.

/*
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 3 of the License, or
 * (at your option) any later version.
 *
 * Written (W) 2012 Chiyuan Zhang
 * Copyright (C) 2012 Chiyuan Zhang
 */

#include <vector>
#include <stack>

#include <shogun/multiclass/tree/ConditionalProbabilityTree.h>
#include <shogun/classifier/svm/OnlineLibLinear.h>

using namespace shogun;
using namespace std;

CMulticlassLabels* CConditionalProbabilityTree::apply_multiclass(CFeatures* data)
{
    if (data)
    {
        if (data->get_feature_class() != C_STREAMING_DENSE)
            SG_ERROR("Expected StreamingDenseFeatures\n");
        if (data->get_feature_type() != F_SHORTREAL)
            SG_ERROR("Expected float32_t feature type\n");

        set_features(dynamic_cast<CStreamingDenseFeatures<float32_t>* >(data));
    }

    vector<int32_t> predicts;

    m_feats->start_parser();
    while (m_feats->get_next_example())
    {
        predicts.push_back(apply_multiclass_example(m_feats->get_vector()));
        m_feats->release_example();
    }
    m_feats->end_parser();

    CMulticlassLabels *labels = new CMulticlassLabels(predicts.size());
    for (size_t i=0; i < predicts.size(); ++i)
        labels->set_int_label(i, predicts[i]);
    return labels;
}

int32_t CConditionalProbabilityTree::apply_multiclass_example(SGVector<float32_t> ex)
{
    compute_conditional_probabilities(ex);
    SGVector<float64_t> probs(m_leaves.size());
    for (map<int32_t,node_t*>::iterator it = m_leaves.begin(); it != m_leaves.end(); ++it)
    {
        probs[it->first] = accumulate_conditional_probability(it->second);
    }
    return SGVector<float64_t>::arg_max(probs.vector, 1, probs.vlen);
}

void CConditionalProbabilityTree::compute_conditional_probabilities(SGVector<float32_t> ex)
{
    stack<node_t *> nodes;
    nodes.push(m_root);

    while (!nodes.empty())
    {
        node_t *node = nodes.top();
        nodes.pop();
        if (node->left())
        {
            nodes.push(node->left());
            nodes.push(node->right());

            // don't calculate for leaf
            node->data.p_right = predict_node(ex, node);
        }
    }
}

float64_t CConditionalProbabilityTree::accumulate_conditional_probability(node_t *leaf)
{
    float64_t prob = 1;
    node_t *par = leaf->parent();
    while (par != NULL)
    {
        if (leaf == par->left())
            prob *= (1-par->data.p_right);
        else
            prob *= par->data.p_right;

        leaf = par;
        par = leaf->parent();
    }

    return prob;
}

bool CConditionalProbabilityTree::train_machine(CFeatures* data)
{
    if (data)
    {
        if (data->get_feature_class() != C_STREAMING_DENSE)
            SG_ERROR("Expected StreamingDenseFeatures\n");
        if (data->get_feature_type() != F_SHORTREAL)
            SG_ERROR("Expected float32_t features\n");
        set_features(dynamic_cast<CStreamingDenseFeatures<float32_t> *>(data));
    }
    else
    {
        if (!m_feats)
            SG_ERROR("No data features provided\n");
    }

    m_machines->reset_array();
    SG_UNREF(m_root);
    m_root = NULL;

    m_leaves.clear();

    m_feats->start_parser();
    for (int32_t ipass=0; ipass < m_num_passes; ++ipass)
    {
        while (m_feats->get_next_example())
        {
            train_example(m_feats->get_vector(), static_cast<int32_t>(m_feats->get_label()));
            m_feats->release_example();
        }

        if (ipass < m_num_passes-1)
            m_feats->reset_stream();
    }
    m_feats->end_parser();

    for (int32_t i=0; i < m_machines->get_num_elements(); ++i)
    {
        COnlineLibLinear *lll = dynamic_cast<COnlineLibLinear *>(m_machines->get_element(i));
        lll->stop_train();
        SG_UNREF(lll);
    }

    return true;
}

void CConditionalProbabilityTree::print_tree()
{
    if (m_root)
        m_root->debug_print(ConditionalProbabilityTreeNodeData::print_data);
    else
        printf("Empty Tree\n");
}

void CConditionalProbabilityTree::train_example(SGVector<float32_t> ex, int32_t label)
{
    if (m_root == NULL)
    {
        m_root = new node_t();
        m_root->data.label = label;
        m_leaves.insert(make_pair(label, m_root));
        m_root->machine(create_machine(ex));
        return;
    }

    if (m_leaves.find(label) != m_leaves.end())
    {
        train_path(ex, m_leaves[label]);
    }
    else
    {
        node_t *node = m_root;
        while (node->left() != NULL)
        {
            // not a leaf
            bool is_left = which_subtree(node, ex);
            float64_t node_label;
            if (is_left)
                node_label = 0;
            else
                node_label = 1;
            train_node(ex, node_label, node);

            if (is_left)
                node = node->left();
            else
                node = node->right();
        }

        m_leaves.erase(node->data.label);

        node_t *left_node = new node_t();
        left_node->data.label = node->data.label;
        node->data.label = -1;
        COnlineLibLinear *node_mch = dynamic_cast<COnlineLibLinear *>(m_machines->get_element(node->machine()));
        COnlineLibLinear *mch = new COnlineLibLinear(node_mch);
        SG_UNREF(node_mch);
        mch->start_train();
        m_machines->push_back(mch);
        left_node->machine(m_machines->get_num_elements()-1);
        m_leaves.insert(make_pair(left_node->data.label, left_node));
        node->left(left_node);

        node_t *right_node = new node_t();
        right_node->data.label = label;
        right_node->machine(create_machine(ex));
        m_leaves.insert(make_pair(label, right_node));
        node->right(right_node);
    }
}

void CConditionalProbabilityTree::train_path(SGVector<float32_t> ex, node_t *node)
{
    float64_t node_label = 0;
    train_node(ex, node_label, node);

    node_t *par = node->parent();
    while (par != NULL)
    {
        if (par->left() == node)
            node_label = 0;
        else
            node_label = 1;

        train_node(ex, node_label, par);
        node = par;
        par = node->parent();
    }
}

void CConditionalProbabilityTree::train_node(SGVector<float32_t> ex, float64_t label, node_t *node)
{
    COnlineLibLinear *mch = dynamic_cast<COnlineLibLinear *>(m_machines->get_element(node->machine()));
    ASSERT(mch);
    mch->train_one(ex, label);
    SG_UNREF(mch);
}

float64_t CConditionalProbabilityTree::predict_node(SGVector<float32_t> ex, node_t *node)
{
    COnlineLibLinear *mch = dynamic_cast<COnlineLibLinear *>(m_machines->get_element(node->machine()));
    ASSERT(mch);
    float64_t pred = mch->apply_one(ex.vector, ex.vlen);
    SG_UNREF(mch);
    // use sigmoid function to turn the decision value into valid probability
    return 1.0/(1+CMath::exp(-pred));
}

int32_t CConditionalProbabilityTree::create_machine(SGVector<float32_t> ex)
{
    COnlineLibLinear *mch = new COnlineLibLinear();
    mch->start_train();
    mch->train_one(ex, 0);
    m_machines->push_back(mch);
    return m_machines->get_num_elements()-1;
}