GUIHMM.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) 1999-2008 Soeren Sonnenburg
 * Written (W) 1999-2008 Gunnar Raetsch
 * Copyright (C) 1999-2008 Fraunhofer Institute FIRST and Max-Planck-Society
 */

#include <shogun/ui/GUIHMM.h>
#include <shogun/ui/SGInterface.h>

#include <shogun/lib/config.h>
#include <shogun/lib/common.h>
#include <shogun/features/StringFeatures.h>
#include <shogun/labels/Labels.h>
#include <shogun/labels/RegressionLabels.h>
#include <shogun/mathematics/Statistics.h>

#include <unistd.h>

using namespace shogun;

CGUIHMM::CGUIHMM(CSGInterface* ui_)
: CSGObject(), ui(ui_)
{
    working=NULL;

    pos=NULL;
    neg=NULL;
    test=NULL;

    PSEUDO=1e-10;
    M=4;
}

CGUIHMM::~CGUIHMM()
{
    SG_UNREF(working);
}

bool CGUIHMM::new_hmm(int32_t n, int32_t m)
{
    SG_UNREF(working);
    working=new CHMM(n, m, NULL, PSEUDO);
    M=m;
    return true;
}

bool CGUIHMM::baum_welch_train()
{
    if (!working)
        SG_ERROR("Create HMM first.\n");

    CFeatures* trainfeatures=ui->ui_features->get_train_features();
    if (!trainfeatures)
        SG_ERROR("Assign train features first.\n");
    if (trainfeatures->get_feature_type()!=F_WORD ||
        trainfeatures->get_feature_class()!=C_STRING)
        SG_ERROR("Features must be STRING of type WORD.\n");

    CStringFeatures<uint16_t>* sf=(CStringFeatures<uint16_t>*) trainfeatures;
    SG_DEBUG("Stringfeatures have %ld orig_symbols %ld symbols %d order %ld max_symbols\n",  (int64_t) sf->get_original_num_symbols(), (int64_t) sf->get_num_symbols(), sf->get_order(), (int64_t) sf->get_max_num_symbols());

    working->set_observations(sf);

    return working->baum_welch_viterbi_train(BW_NORMAL);
}


bool CGUIHMM::baum_welch_trans_train()
{
    if (!working)
        SG_ERROR("Create HMM first.\n");

    CFeatures* trainfeatures=ui->ui_features->get_train_features();
    if (!trainfeatures)
        SG_ERROR("Assign train features first.\n");
    if (trainfeatures->get_feature_type()!=F_WORD ||
        trainfeatures->get_feature_class()!=C_STRING)
        SG_ERROR("Features must be STRING of type WORD.\n");

    working->set_observations((CStringFeatures<uint16_t>*) trainfeatures);

    return working->baum_welch_viterbi_train(BW_TRANS);
}


bool CGUIHMM::baum_welch_train_defined()
{
    if (!working)
        SG_ERROR("Create HMM first.\n");
    if (!working->get_observations())
        SG_ERROR("Assign observation first.\n");

    return working->baum_welch_viterbi_train(BW_DEFINED);
}

bool CGUIHMM::viterbi_train()
{
    if (!working)
        SG_ERROR("Create HMM first.\n");
    if (!working->get_observations())
        SG_ERROR("Assign observation first.\n");

    return working->baum_welch_viterbi_train(VIT_NORMAL);
}

bool CGUIHMM::viterbi_train_defined()
{
    if (!working)
        SG_ERROR("Create HMM first.\n");
    if (!working->get_observations())
        SG_ERROR("Assign observation first.\n");

    return working->baum_welch_viterbi_train(VIT_DEFINED);
}

bool CGUIHMM::linear_train(char align)
{
    if (!working)
        SG_ERROR("Create HMM first.\n");

    CFeatures* trainfeatures=ui->ui_features->get_train_features();
    if (!trainfeatures)
        SG_ERROR("Assign train features first.\n");
    if (trainfeatures->get_feature_type()!=F_WORD ||
        trainfeatures->get_feature_class()!=C_STRING)
        SG_ERROR("Features must be STRING of type WORD.\n");

    working->set_observations((CStringFeatures<uint16_t>*) ui->
        ui_features->get_train_features());

    bool right_align=false;
    if (align=='r')
    {
        SG_INFO("Using alignment to right.\n");
        right_align=true;
    }
    else
        SG_INFO("Using alignment to left.\n");
    working->linear_train(right_align);

    return true;
}

CRegressionLabels* CGUIHMM::classify(CRegressionLabels* result)
{
    CStringFeatures<uint16_t>* obs= (CStringFeatures<uint16_t>*) ui->
        ui_features->get_test_features();
    ASSERT(obs);
    int32_t num_vec=obs->get_num_vectors();

    //CStringFeatures<uint16_t>* old_pos=pos->get_observations();
    //CStringFeatures<uint16_t>* old_neg=neg->get_observations();

    pos->set_observations(obs);
    neg->set_observations(obs);

    if (!result)
        result=new CRegressionLabels(num_vec);

    for (int32_t i=0; i<num_vec; i++)
        result->set_label(i, pos->model_probability(i) - neg->model_probability(i));

    //pos->set_observations(old_pos);
    //neg->set_observations(old_neg);
    return result;
}

float64_t CGUIHMM::classify_example(int32_t idx)
{
    CStringFeatures<uint16_t>* obs= (CStringFeatures<uint16_t>*) ui->
        ui_features->get_test_features();
    ASSERT(obs);

    //CStringFeatures<uint16_t>* old_pos=pos->get_observations();
    //CStringFeatures<uint16_t>* old_neg=neg->get_observations();

    pos->set_observations(obs);
    neg->set_observations(obs);

    float64_t result=pos->model_probability(idx) - neg->model_probability(idx);
    //pos->set_observations(old_pos);
    //neg->set_observations(old_neg);
    return result;
}

CRegressionLabels* CGUIHMM::one_class_classify(CRegressionLabels* result)
{
    ASSERT(working);

    CStringFeatures<uint16_t>* obs= (CStringFeatures<uint16_t>*) ui->
        ui_features->get_test_features();
    ASSERT(obs);
    int32_t num_vec=obs->get_num_vectors();

    //CStringFeatures<uint16_t>* old_pos=working->get_observations();
    working->set_observations(obs);

    if (!result)
        result=new CRegressionLabels(num_vec);

    for (int32_t i=0; i<num_vec; i++)
        result->set_label(i, working->model_probability(i));

    //working->set_observations(old_pos);
    return result;
}

CRegressionLabels* CGUIHMM::linear_one_class_classify(CRegressionLabels* result)
{
    ASSERT(working);

    CStringFeatures<uint16_t>* obs= (CStringFeatures<uint16_t>*) ui->
        ui_features->get_test_features();
    ASSERT(obs);
    int32_t num_vec=obs->get_num_vectors();

    //CStringFeatures<uint16_t>* old_pos=working->get_observations();
    working->set_observations(obs);

    if (!result)
        result=new CRegressionLabels(num_vec);

    for (int32_t i=0; i<num_vec; i++)
        result->set_label(i, working->linear_model_probability(i));

    //working->set_observations(old_pos);
    return result;
}


float64_t CGUIHMM::one_class_classify_example(int32_t idx)
{
    ASSERT(working);

    CStringFeatures<uint16_t>* obs= (CStringFeatures<uint16_t>*) ui->
        ui_features->get_test_features();
    ASSERT(obs);

    //CStringFeatures<uint16_t>* old_pos=pos->get_observations();

    pos->set_observations(obs);
    neg->set_observations(obs);

    float64_t result=working->model_probability(idx);
    //working->set_observations(old_pos);
    return result;
}

bool CGUIHMM::append_model(char* filename, int32_t base1, int32_t base2)
{
    if (!working)
        SG_ERROR("Create HMM first.\n");
    if (!filename)
        SG_ERROR("Invalid filename.\n");

    FILE* model_file=fopen(filename, "r");
    if (!model_file)
        SG_ERROR("Opening file %s failed.\n", filename);

    CHMM* h=new CHMM(model_file,PSEUDO);
    if (!h || !h->get_status())
    {
        SG_UNREF(h);
        fclose(model_file);
        SG_ERROR("Reading file %s failed.\n", filename);
    }

    fclose(model_file);
    SG_INFO("File %s successfully read.\n", filename);

    SG_DEBUG("h %d , M: %d\n", h, h->get_M());
    if (base1!=-1 && base2!=-1)
    {
        float64_t* cur_o=SG_MALLOC(float64_t, h->get_M());
        float64_t* app_o=SG_MALLOC(float64_t, h->get_M());

        for (int32_t i=0; i<h->get_M(); i++)
        {
            if (i==base1)
                cur_o[i]=0;
            else
                cur_o[i]=-1000;

            if (i==base2)
                app_o[i]=0;
            else
                app_o[i]=-1000;
        }

        working->append_model(h, cur_o, app_o);

        SG_FREE(cur_o);
        SG_FREE(app_o);
    }
    else
        working->append_model(h);

    SG_UNREF(h);
    SG_INFO("New model has %i states.\n", working->get_N());
    return true;
}

bool CGUIHMM::add_states(int32_t num_states, float64_t value)
{
    if (!working)
        SG_ERROR("Create HMM first.\n");

    working->add_states(num_states, value);
    SG_INFO("New model has %i states, value %f.\n", working->get_N(), value);
    return true;
}

bool CGUIHMM::set_pseudo(float64_t pseudo)
{
    PSEUDO=pseudo;
    SG_INFO("Current setting: pseudo=%e.\n", PSEUDO);
    return true;
}

bool CGUIHMM::convergence_criteria(int32_t num_iterations, float64_t epsilon)
{
    if (!working)
        SG_ERROR("Create HMM first.\n");

    working->set_iterations(num_iterations);
    working->set_epsilon(epsilon);

    SG_INFO("Current HMM convergence criteria: iterations=%i, epsilon=%e\n", working->get_iterations(), working->get_epsilon());
    return true;
}

bool CGUIHMM::set_hmm_as(char* target)
{
    if (!working)
        SG_ERROR("Create HMM first!\n");

    if (strncmp(target, "POS", 3)==0)
    {
        SG_UNREF(pos);
        pos=working;
        working=NULL;
    }
    else if (strncmp(target, "NEG", 3)==0)
    {
        SG_UNREF(neg);
        neg=working;
        working=NULL;
    }
    else if (strncmp(target, "TEST", 4)==0)
    {
        SG_UNREF(test);
        test=working;
        working=NULL;
    }
    else
        SG_ERROR("Target POS|NEG|TEST is missing.\n");

    return true;
}

bool CGUIHMM::load(char* filename)
{
    bool result=false;

    FILE* model_file=fopen(filename, "r");
    if (!model_file)
        SG_ERROR("Opening file %s failed.\n", filename);

    SG_UNREF(working);
    working=new CHMM(model_file, PSEUDO);
    fclose(model_file);

    if (working && working->get_status())
    {
        SG_INFO("Loaded HMM successfully from file %s.\n", filename);
        result=true;
    }

    M=working->get_M();

    return result;
}

bool CGUIHMM::save(char* filename, bool is_binary)
{
    bool result=false;

    if (!working)
        SG_ERROR("Create HMM first.\n");

    FILE* file=fopen(filename, "w");
    if (file)
    {
        if (is_binary)
            result=working->save_model_bin(file);
        else
            result=working->save_model(file);
    }

    if (!file || !result)
        SG_ERROR("Writing to file %s failed!\n", filename);
    else
        SG_INFO("Successfully written model into %s!\n", filename);

    if (file)
        fclose(file);

    return result;
}

bool CGUIHMM::load_definitions(char* filename, bool do_init)
{
    if (!working)
        SG_ERROR("Create HMM first.\n");

    bool result=false;
    FILE* def_file=fopen(filename, "r");
    if (!def_file)
        SG_ERROR("Opening file %s failed\n", filename);

    if (working->load_definitions(def_file, true, do_init))
    {
        SG_INFO("Definitions successfully read from %s.\n", filename);
        result=true;
    }
    else
        SG_ERROR("Couldn't load definitions form file %s.\n", filename);

    fclose(def_file);
    return result;
}

bool CGUIHMM::save_likelihood(char* filename, bool is_binary)
{
    bool result=false;

    if (!working)
        SG_ERROR("Create HMM first\n");

    FILE* file=fopen(filename, "w");
    if (file)
    {
        //if (binary)
        //  result=working->save_model_bin(file);
        //else

        result=working->save_likelihood(file);
    }

    if (!file || !result)
        SG_ERROR("Writing to file %s failed!\n", filename);
    else
        SG_INFO("Successfully written likelihoods into %s!\n", filename);

    if (file)
        fclose(file);

    return result;
}

bool CGUIHMM::save_path(char* filename, bool is_binary)
{
    bool result=false;
    if (!working)
        SG_ERROR("Create HMM first.\n");

    FILE* file=fopen(filename, "w");
    if (file)
    {
        //if (binary)
        //_train()/ result=working->save_model_bin(file);
        //else
        CStringFeatures<uint16_t>* obs=(CStringFeatures<uint16_t>*) ui->
            ui_features->get_test_features();
        ASSERT(obs);
        working->set_observations(obs);

        result=working->save_path(file);
    }

    if (!file || !result)
        SG_ERROR("Writing to file %s failed!\n", filename);
    else
        SG_INFO("Successfully written path into %s!\n", filename);

    if (file)
        fclose(file);

    return result;
}

bool CGUIHMM::chop(float64_t value)
{
    if (!working)
        SG_ERROR("Create HMM first.\n");

    working->chop(value);
    return true;
}

bool CGUIHMM::likelihood()
{
    if (!working)
        SG_ERROR("Create HMM first!\n");

    working->output_model(false);
    return true;
}

bool CGUIHMM::output_hmm()
{
    if (!working)
        SG_ERROR("Create HMM first!\n");

    working->output_model(true);
    return true;
}

bool CGUIHMM::output_hmm_defined()
{
    if (!working)
        SG_ERROR("Create HMM first!\n");

    working->output_model_defined(true);
    return true;
}

bool CGUIHMM::best_path(int32_t from, int32_t to)
{
    // FIXME: from unused???

    if (!working)
        SG_ERROR("Create HMM first.\n");

    //get path
    working->best_path(0);

    for (int32_t t=0; t<working->get_observations()->get_vector_length(0)-1 && t<to; t++)
        SG_PRINT("%d ", working->get_best_path_state(0, t));
    SG_PRINT("\n");

    //for (t=0; t<p_observations->get_vector_length(0)-1 && t<to; t++)
    //  SG_PRINT( "%d ", PATH(0)[t]);
    //
    return true;
}

bool CGUIHMM::normalize(bool keep_dead_states)
{
    if (!working)
        SG_ERROR("Create HMM first.\n");

    working->normalize(keep_dead_states);
    return true;
}

bool CGUIHMM::relative_entropy(float64_t*& values, int32_t& len)
{
    if (!pos || !neg)
        SG_ERROR("Set pos and neg HMM first!\n");

    int32_t pos_N=pos->get_N();
    int32_t neg_N=neg->get_N();
    int32_t pos_M=pos->get_M();
    int32_t neg_M=neg->get_M();
    if (pos_M!=neg_M || pos_N!=neg_N)
        SG_ERROR("Pos and neg HMM's differ in number of emissions or states.\n");

    float64_t* p=SG_MALLOC(float64_t, pos_M);
    float64_t* q=SG_MALLOC(float64_t, neg_M);

    SG_FREE(values);
    values=SG_MALLOC(float64_t, pos_N);

    for (int32_t i=0; i<pos_N; i++)
    {
        for (int32_t j=0; j<pos_M; j++)
        {
            p[j]=pos->get_b(i, j);
            q[j]=neg->get_b(i, j);
        }

        values[i]=CStatistics::relative_entropy(p, q, pos_M);
    }
    SG_FREE(p);
    SG_FREE(q);

    len=pos_N;
    return true;
}

bool CGUIHMM::entropy(float64_t*& values, int32_t& len)
{
    if (!working)
        SG_ERROR("Create HMM first!\n");

    int32_t n=working->get_N();
    int32_t m=working->get_M();
    float64_t* p=SG_MALLOC(float64_t, m);

    SG_FREE(values);
    values=SG_MALLOC(float64_t, n);

    for (int32_t i=0; i<n; i++)
    {
        for (int32_t j=0; j<m; j++)
            p[j]=working->get_b(i, j);

        values[i]=CStatistics::entropy(p, m);
    }
    SG_FREE(p);

    len=m;
    return true;
}

bool CGUIHMM::permutation_entropy(int32_t width, int32_t seq_num)
{
    if (!working)
        SG_ERROR("Create hmm first.\n");

    if (!working->get_observations())
        SG_ERROR("Set observations first.\n");

    return working->permutation_entropy(width, seq_num);
}