root/library/bdm/base/bdmbase.cpp @ 746

#include "bdmbase.h"

//! Space of basic BDM structures
namespace bdm {

const int RV::BUFFER_STEP = 1;

Array<string> RV::NAMES ( RV::BUFFER_STEP );

ivec RV::SIZES ( RV::BUFFER_STEP );

RV::str2int_map RV::MAP;

void RV::clear_all() {
        MAP.clear();
        SIZES.clear();
        NAMES = Array<string> ( BUFFER_STEP );
}

string RV::show_all() {
        ostringstream os;
        for ( str2int_map::const_iterator iter = MAP.begin(); iter != MAP.end(); iter++ ) {
                os << "key: " << iter->first << " val: " << iter->second << endl;
        }
        return os.str();
};

int RV::init ( const string &name, int size ) {
        //Refer
        int id;
        str2int_map::const_iterator iter = MAP.find ( name );
        if ( iter == MAP.end() || name.length() == 0 ) { //add new RV
                id = MAP.size() + 1;
                //debug
                /*              {
                                        cout << endl;
                                        str2int_map::const_iterator iter = MAP.begin();
                                        for(str2int_map::const_iterator iter=MAP.begin(); iter!=MAP.end(); iter++){
                                                cout << "key: " << iter->first << " val: " << iter->second <<endl;
                                        }
                                }*/

                MAP.insert ( make_pair ( name, id ) ); //add new rv
                if ( id >= NAMES.length() ) {
                        NAMES.set_length ( id + BUFFER_STEP, true );
                        SIZES.set_length ( id + BUFFER_STEP, true );
                }
                NAMES ( id ) = name;
                SIZES ( id ) = size;
                bdm_assert ( size > 0, "RV " + name + " does not exists. Default size (-1) can not be assigned " );
        } else {
                id = iter->second;
                if ( size > 0 && name.length() > 0 ) {
                        bdm_assert ( SIZES ( id ) == size, "RV " + name + " of size " + num2str ( SIZES ( id ) ) + " exists, requested size " + num2str ( size ) + "can not be assigned" );
                }
        }
        return id;
};

int RV::countsize() const {
        int tmp = 0;
        for ( int i = 0; i < len; i++ ) {
                tmp += SIZES ( ids ( i ) );
        }
        return tmp;
}

ivec RV::cumsizes() const {
        ivec szs ( len );
        int tmp = 0;
        for ( int i = 0; i < len; i++ ) {
                tmp += SIZES ( ids ( i ) );
                szs ( i ) = tmp;
        }
        return szs;
}

void RV::init ( const Array<std::string> &in_names, const ivec &in_sizes, const ivec &in_times ) {
        len = in_names.length();
        bdm_assert ( in_names.length() == in_times.length(), "check \"times\" " );
        bdm_assert ( in_names.length() == in_sizes.length(), "check \"sizes\" " );

        times.set_length ( len );
        ids.set_length ( len );
        int id;
        for ( int i = 0; i < len; i++ ) {
                id = init ( in_names ( i ), in_sizes ( i ) );
                ids ( i ) = id;
        }
        times = in_times;
        dsize = countsize();
}

RV::RV ( string name, int sz, int tm ) {
        Array<string> A ( 1 );
        A ( 0 ) = name;
        init ( A, vec_1 ( sz ), vec_1 ( tm ) );
}

bool RV::add ( const RV &rv2 ) {
        // TODO
        if ( rv2.len > 0 ) { //rv2 is nonempty
                ivec ind = rv2.findself ( *this ); //should be -1 all the time
                ivec index = itpp::find ( ind == -1 );

                if ( index.length() < rv2.len ) { //conflict
                        ids = concat ( ids, rv2.ids ( index ) );
                        times = concat ( times, rv2.times ( index ) );
                } else {
                        ids = concat ( ids, rv2.ids );
                        times = concat ( times, rv2.times );
                }
                len = ids.length();
                dsize = countsize();
                return ( index.length() == rv2.len ); //conflict or not
        } else { //rv2 is empty
                return true; // no conflict
        }
};

RV RV::subselect ( const ivec &ind ) const {
        RV ret;
        ret.ids = ids ( ind );
        ret.times = times ( ind );
        ret.len = ind.length();
        ret.dsize = ret.countsize();
        return ret;
}

RV RV::operator() ( int di1, int di2 ) const {
        ivec sz = cumsizes();
        int i1 = 0;
        while ( sz ( i1 ) < di1 ) i1++;
        int i2 = i1;
        while ( sz ( i2 ) < di2 ) i2++;
        return subselect ( linspace ( i1, i2 ) );
}

void RV::t_plus ( int delta ) {
        times += delta;
}

bool RV::equal ( const RV &rv2 ) const {
        return ( ids == rv2.ids ) && ( times == rv2.times );
}

shared_ptr<pdf> epdf::condition ( const RV &rv ) const {
        bdm_warning ( "Not implemented" );
        return shared_ptr<pdf>();
}

shared_ptr<epdf> epdf::marginal ( const RV &rv ) const {
        bdm_warning ( "Not implemented" );
        return shared_ptr<epdf>();
}

mat epdf::sample_mat ( int N ) const {
        mat X = zeros ( dim, N );
        for ( int i = 0; i < N; i++ ) X.set_col ( i, this->sample() );
        return X;
}

vec epdf::evallog_mat ( const mat &Val ) const {
        vec x ( Val.cols() );
        for ( int i = 0; i < Val.cols(); i++ ) {
                x ( i ) = evallog ( Val.get_col ( i ) );
        }

        return x;
}

vec epdf::evallog_mat ( const Array<vec> &Avec ) const {
        vec x ( Avec.size() );
        for ( int i = 0; i < Avec.size(); i++ ) {
                x ( i ) = evallog ( Avec ( i ) );
        }

        return x;
}

mat pdf::samplecond_mat ( const vec &cond, int N ) {
        mat M ( dimension(), N );
        for ( int i = 0; i < N; i++ ) {
                M.set_col ( i, samplecond ( cond ) );
        }

        return M;
}

void pdf::from_setting ( const Setting &set ) {
        root::from_setting( set );
        shared_ptr<RV> r = UI::build<RV> ( set, "rv", UI::optional );
        if ( r ) {
                set_rv ( *r );
        }

        r = UI::build<RV> ( set, "rvc", UI::optional );
        if ( r ) {
                set_rvc ( *r );
        }
}

void pdf::to_setting ( Setting &set ) const {   
        root::to_setting( set );
        UI::save( &rv, set, "rv" );
        UI::save( &rvc, set, "rvc" );
}

void datalink::set_connection ( const RV &rv, const RV &rv_up ) {
        downsize = rv._dsize();
        upsize = rv_up._dsize();
        v2v_up = rv.dataind ( rv_up );
        bdm_assert_debug ( v2v_up.length() == downsize, "rv is not fully in rv_up" );
}

void datalink::set_connection ( int ds, int us, const ivec &upind ) {
        downsize = ds;
        upsize = us;
        v2v_up = upind;
        bdm_assert_debug ( v2v_up.length() == downsize, "rv is not fully in rv_up" );
}

void datalink_part::set_connection ( const RV &rv, const RV &rv_up ) {
        rv.dataind ( rv_up, v2v_down, v2v_up );
        downsize = v2v_down.length();
        upsize = v2v_up.length();
}

void datalink_m2e::set_connection ( const RV &rv, const RV &rvc, const RV &rv_up ) {
        datalink::set_connection ( rv, rv_up );
        condsize = rvc._dsize();
        //establish v2c connection
        rvc.dataind ( rv_up, v2c_lo, v2c_up );
}

vec datalink_m2e::get_cond ( const vec &val_up ) {
        vec tmp ( condsize );
        set_subvector ( tmp, v2c_lo, val_up ( v2c_up ) );
        return tmp;
}

void datalink_m2e::pushup_cond ( vec &val_up, const vec &val, const vec &cond ) {
        bdm_assert_debug ( downsize == val.length(), "Wrong val" );
        bdm_assert_debug ( upsize == val_up.length(), "Wrong val_up" );
        set_subvector ( val_up, v2v_up, val );
        set_subvector ( val_up, v2c_up, cond );
}

std::ostream &operator<< ( std::ostream &os, const RV &rv ) {
        int id;
        for ( int i = 0; i < rv.len ; i++ ) {
                id = rv.ids ( i );
                os << id << "(" << RV::SIZES ( id ) << ")" <<  // id(size)=
                "=" << RV::NAMES ( id )  << "_{"  << rv.times ( i ) << "}; "; //name_{time}
        }
        return os;
}

RV RV::expand_delayes() const {
        RV rvt = this->remove_time(); //rv at t=0
        RV tmp = rvt;
        int td = mint();
        for ( int i = -1; i >= td; i-- ) {
                rvt.t_plus ( -1 );
                tmp.add ( rvt ); //shift u1
        }
        return tmp;
}

str RV::tostr() const {
        ivec idlist ( dsize );
        ivec tmlist ( dsize );
        int i;
        int pos = 0;
        for ( i = 0; i < len; i++ ) {
                idlist.set_subvector ( pos, pos + size ( i ) - 1, ids ( i ) );
                tmlist.set_subvector ( pos, pos + size ( i ) - 1, times ( i ) );
                pos += size ( i );
        }
        return str ( idlist, tmlist );
}

ivec RV::dataind ( const RV &rv2 ) const {
        ivec res ( 0 );
        if ( rv2._dsize() > 0 ) {
                str str2 = rv2.tostr();
                ivec part;
                int i;
                for ( i = 0; i < len; i++ ) {
                        part = itpp::find ( ( str2.ids == ids ( i ) ) & ( str2.times == times ( i ) ) );
                        res = concat ( res, part );
                }
        }

        bdm_assert_debug ( res.length() == dsize, "this rv is not fully present in crv!" );
        return res;

}

void RV::dataind ( const RV &rv2, ivec &selfi, ivec &rv2i ) const {
        //clean results
        selfi.set_size ( 0 );
        rv2i.set_size ( 0 );

        // just in case any rv is empty
        if ( ( len == 0 ) || ( rv2.length() == 0 ) ) {
                return;
        }

        //find comon rv
        ivec cids = itpp::find ( this->findself ( rv2 ) >= 0 );

        // index of
        if ( cids.length() > 0 ) {
                str str1 = tostr();
                str str2 = rv2.tostr();

                ivec part1;
                ivec part2;
                int i, j;
                // find common rv in strs
                for ( j = 0; j < cids.length(); j++ ) {
                        i = cids ( j );
                        part1 = itpp::find ( ( str1.ids == ids ( i ) ) & ( str1.times == times ( i ) ) );
                        part2 = itpp::find ( ( str2.ids == ids ( i ) ) & ( str2.times == times ( i ) ) );
                        selfi = concat ( selfi, part1 );
                        rv2i = concat ( rv2i, part2 );
                }
        }
        bdm_assert_debug ( selfi.length() == rv2i.length(), "this should not happen!" );
}

RV RV::subt ( const RV &rv2 ) const {
        ivec res = this->findself ( rv2 ); // nonzeros
        ivec valid;
        if ( dsize > 0 ) {
                valid = itpp::find ( res == -1 );    //-1 => value not found => it remains
        }
        return ( *this ) ( valid ); //keep those that were not found in rv2
}

std::string RV::scalarname ( int scalat ) const {
        bdm_assert ( scalat < dsize, "Wrong input index" );
        int id = 0;
        int scalid = 0;
        while ( scalid + SIZES ( ids ( id ) ) <= scalat )  {
                scalid += SIZES ( ids ( id ) );
                id++;
        };
        //now id is the id of variable of interest
        if ( size ( id ) == 1 )
                return  NAMES ( ids ( id ) );
        else
                return  NAMES ( ids ( id ) ) + "_" + num2str ( scalat - scalid );

}

ivec RV::findself ( const RV &rv2 ) const {
        int i, j;
        ivec tmp = -ones_i ( len );
        for ( i = 0; i < len; i++ ) {
                for ( j = 0; j < rv2.length(); j++ ) {
                        if ( ( ids ( i ) == rv2.ids ( j ) ) & ( times ( i ) == rv2.times ( j ) ) ) {
                                tmp ( i ) = j;
                                break;
                        }
                }
        }
        return tmp;
}

ivec RV::findself_ids ( const RV &rv2 ) const {
        int i, j;
        ivec tmp = -ones_i ( len );
        for ( i = 0; i < len; i++ ) {
                for ( j = 0; j < rv2.length(); j++ ) {
                        if ( ( ids ( i ) == rv2.ids ( j ) ) ) {
                                tmp ( i ) = j;
                                break;
                        }
                }
        }
        return tmp;
}

void RV::from_setting ( const Setting &set ) {
        Array<string> A;
        UI::get ( A, set, "names" );

        ivec szs;
        if ( !UI::get ( szs, set, "sizes" ) )
                szs = ones_i ( A.length() );

        ivec tms;
        if ( !UI::get ( tms, set, "times" ) )
                tms = zeros_i ( A.length() );

        // TODO tady se bude plnit primo do jeho promennych, a pak se zavola validacnni metoda, takze cele prepsat, ano?
        init ( A, szs, tms );
}

void RV::to_setting ( Setting &set ) const {
        Array<string> names ( len );
        ivec sizes ( len );
        for ( int i = 0; i < len; i++ ) {
                names ( i ) = name ( i );
                sizes ( i ) = size ( i );
        }
        UI::save ( names, set, "names" );
        UI::save ( sizes, set, "sizes" );
        UI::save ( times, set, "times" );
}

RV concat ( const RV &rv1, const RV &rv2 ) {
        RV pom = rv1;
        pom.add ( rv2 );
        return pom;
}

RV get_composite_rv ( const Array<shared_ptr<pdf> > &pdfs,
                      bool checkoverlap ) {
        RV rv; //empty rv
        bool rvaddok;
        for ( int i = 0; i < pdfs.length(); i++ ) {
                rvaddok = rv.add ( pdfs ( i )->_rv() ); //add rv to common rvs.
                // If rvaddok==false, pdfs overlap => assert error.
                bdm_assert_debug ( rvaddok || !checkoverlap, "mprod::mprod() input pdfs overlap in rv!" );
        }

        return rv;
}

int logger::add_vector ( const RV &rv, string prefix ) {
        int id;
        if ( rv._dsize() > 0 ) {
                id = entries.length();
                names = concat ( names, prefix ); // diff
                entries.set_length ( id + 1, true );
                entries ( id ) = rv;
        } else {
                id = -1;
        }
        return id; // identifier of the last entry
}

int logger::add_setting ( const string &prefix ) {
        Setting &root = setting_conf.getRoot();
        int id = root.getLength(); //root must be group!!
        if ( prefix.length() > 0 ) {
                settings.set_length ( id + 1, true );
                settings ( id ) = &root.add ( prefix, Setting::TypeList );
        } else {
                id = -1;
        }
        return id;
}

void epdf::log_register ( logger &L, const string &prefix ) {
        RV r;
        if ( isnamed() ) {
                r = _rv();
        } else {
                r = RV ( "", dimension() );
        };
        root::log_register ( L, prefix );

        // log full data
        if ( log_level == 10 ) {
                logrec->ids.set_size ( 1 );
                logrec->ids ( 0 ) = logrec->L.add_setting ( prefix );
        } else {
                // log only
                logrec->ids.set_size ( 3 );
                if ( log_level > 0 ) {
                        logrec->ids ( 0 ) = logrec->L.add_vector ( r, prefix + logrec->L.prefix_sep() + "mean" );
                }
                if ( log_level > 1 ) {
                        logrec->ids ( 1 ) = logrec->L.add_vector ( r, prefix + logrec->L.prefix_sep() + "lb" );
                        logrec->ids ( 2 ) = logrec->L.add_vector ( r, prefix + logrec->L.prefix_sep() + "ub" );
                }
        }
}

void epdf::log_write() const {
        if ( log_level == 10 ) {
                to_setting ( logrec->L.log_to_setting ( logrec->ids ( 0 ) ) );
        } else {
                if ( log_level > 0 ) {
                        logrec->L.log_vector ( logrec->ids ( 0 ), mean() );
                }
                if ( log_level > 1 ) {
                        vec lb;
                        vec ub;
                        qbounds ( lb, ub );
                        logrec->L.log_vector ( logrec->ids ( 1 ), lb );
                        logrec->L.log_vector ( logrec->ids ( 2 ), ub );
                }
        }
}

void datalink_buffered::set_connection ( const RV &rv, const RV &rv_up ) {
        // create link between up and down
        datalink_part::set_connection ( rv, rv_up); // only non-delayed version

        // create rvs of history
        // we can store only what we get in rv_up - everything else is removed
        ivec valid_ids = rv.findself_ids ( rv_up ); // return on which position each id is
        RV rv_hist = rv.subselect ( find ( valid_ids >= 0 ) ); // select only rvs that are in rv_up, ie ind>0
        RV rv_hist0 = rv_hist.remove_time(); // these RVs will form history at time =0
        // now we need to know what is needed from Up
        rv_hist = rv_hist.expand_delayes(); // full regressor - including time 0
        Hrv = rv_hist.subt ( rv_hist0 );   // remove time 0
        history = zeros ( Hrv._dsize() );

        // decide if we need to copy val to history
        if ( Hrv._dsize() > 0 ) {
                v2h_up = rv_hist0.dataind ( rv_up ); // indeces of elements of rv_up to be copied
        } // else v2h_up is empty

        Hrv.dataind ( rv, h2v_hist, h2v_down );

        downsize = v2v_down.length() + h2v_down.length();
        upsize = v2v_up.length();
}

void datalink_buffered::set_history ( const RV& rv1, const vec &hist0 ) {
        bdm_assert ( rv1._dsize() == hist0.length(), "hist is not compatible with given rv1" );
        ivec ind_H;
        ivec ind_h0;
        Hrv.dataind ( rv1, ind_H, ind_h0 ); // find indeces of rv in
        set_subvector ( history, ind_H, hist0 ( ind_h0 ) ); // copy given hist to appropriate places
}

void DS::log_register ( logger &L,  const string &prefix ) {
        bdm_assert ( ytsize == Yrv._dsize(), "invalid DS: ytsize (" + num2str ( ytsize ) + ") different from Drv " + num2str ( Yrv._dsize() ) );
        bdm_assert ( utsize == Urv._dsize(), "invalid DS: utsize (" + num2str ( utsize ) + ") different from Urv " + num2str ( Urv._dsize() ) );

        root::log_register ( L, prefix );
        //we know that
        if ( log_level > 0 ) {
                logrec->ids.set_size ( 2 );
                logrec->ids ( 0 ) = logrec->L.add_vector ( Yrv, prefix );
                logrec->ids ( 1 ) = logrec->L.add_vector ( Urv, prefix );
        }
}

void DS::log_write ( ) const {
        if ( log_level > 0 ) {
                vec tmp ( Yrv._dsize() + Urv._dsize() );
                getdata ( tmp );
                // d is first in getdata
                logrec->L.log_vector ( logrec->ids ( 0 ), tmp.left ( Yrv._dsize() ) );
                // u follows after d in getdata
                logrec->L.log_vector ( logrec->ids ( 1 ), tmp.mid ( Yrv._dsize(), Urv._dsize() ) );
        }
}

void BM::set_options ( const string &opt ) {
        if ( opt.find ( "logfull" ) != string::npos ) {
                const_cast<epdf&> ( posterior() ).set_log_level ( 10 ) ;
        } else {
                if ( opt.find ( "logbounds" ) != string::npos ) {
                        const_cast<epdf&> ( posterior() ).set_log_level ( 2 ) ;
                } else {
                        const_cast<epdf&> ( posterior() ).set_log_level ( 1 ) ;
                }
                if ( opt.find ( "logll" ) != string::npos ) {
                        log_level = 1;
                }
        }
}

void BM::log_register ( logger &L, const string &prefix ) {
        root::log_register ( L, prefix );

        const_cast<epdf&> ( posterior() ).log_register ( L, prefix + L.prefix_sep() + "apost" );

        if ( ( log_level ) > 0 ) {
                logrec->ids.set_size ( 1 );
                logrec->ids ( 0 ) = L.add_vector ( RV ( "", 1 ), prefix + L.prefix_sep() + "ll" );
        }
}

void BM::log_write ( ) const {
        posterior().log_write();
        if ( log_level > 0 ) {
                logrec->L.logit ( logrec->ids ( 0 ), ll );
        }
}

void BM::bayes_batch ( const mat &Data, const vec &cond ) {
        for ( int t = 0; t < Data.cols(); t++ ) {
                bayes ( Data.get_col ( t ), cond );
        }
}

void BM::bayes_batch ( const mat &Data, const mat &Cond ) {
        for ( int t = 0; t < Data.cols(); t++ ) {
                bayes ( Data.get_col ( t ), Cond.get_col ( t ) );
        }
}

}