/*! \file \brief Common DataSources. \author Vaclav Smidl. ----------------------------------- BDM++ - C++ library for Bayesian Decision Making under Uncertainty Using IT++ for numerical operations ----------------------------------- */ #ifndef DS_H #define DS_H #include "libBM.h" #include "libEF.h" namespace bdm { /*! * \brief Memory storage of off-line data column-wise The data are stored in an internal matrix \c Data . Each column of Data corresponds to one discrete time observation \f$t\f$. Access to this matrix is via indices \c rowid and \c delays. The data can be loaded from a file. */ class MemDS : public DS { protected: //! internal matrix of data mat Data; //! active column in the Data matrix int time; //! vector of rows that are presented in Dt ivec rowid; //! vector of delays that are presented in Dt ivec delays; public: void getdata ( vec &dt ); void getdata ( vec &dt, const ivec &indeces ); void set_rvs ( RV &drv, RV &urv ); void write ( vec &ut ) {it_error ( "MemDS::write is not supported" );} void write ( vec &ut,ivec &indices ) {it_error ( "MemDS::write is not supported" );} void step(); //!Default constructor MemDS () {}; MemDS ( mat &Dat, ivec &rowid, ivec &delays ); }; /*! Pseudovirtual class for reading data from files */ class FileDS: public MemDS { public: void getdata ( vec &dt ) { it_assert_debug ( dt.length() ==Data.rows(),"" ); dt = Data.get_col(time); }; void getdata ( vec &dt, const ivec &indeces ){ it_assert_debug ( dt.length() ==indeces.length(),"" ); vec tmp(indeces.length()); tmp = Data.get_col(time); dt = tmp(indeces); }; //! returns number of data in the file; int ndat(){return Data.cols();} }; /*! * \brief Read Data Matrix from an IT file The constructor creates an internal matrix \c Data from an IT++ file. The file is binary and can be made using the IT++ library or the Matlab/Octave function itsave. NB: the data are stored columnwise, i.e. each column contains the data for time \f$t\f$! */ class ItppFileDS: public FileDS { public: ItppFileDS ( const string &fname, const string &varname ) :FileDS() { it_file it ( fname ); it << Name ( varname ); it >> Data; time = 0; //rowid and delays are ignored } }; /*! * \brief CSV file data storage The constructor creates \c Data matrix from the records in a CSV file \c fname. The orientation can be of two types: 1. \c BY_COL which is default - the data are stored in columns; one column per time \f$t\f$, one row per data item. 2. \c BY_ROW if the data are stored the classical CSV style. Then each column stores the values for data item, for ex. \f$[y_{t} y_{t-1} ...]\f$, one row for each discrete time instant. */ class CsvFileDS: public FileDS { public: //! Constructor - create DS from a CSV file. CsvFileDS ( const string& fname, const string& orientation = "BY_COL" ); }; /*! \brief Generator of ARX data */ class ArxDS : public DS { protected: //! Rv of the regressor RV Rrv; //! History, ordered as \f$[y_t, u_t, y_{t-1 }, u_{t-1}, \ldots]\f$ vec H; //! (future) input vec U; //! temporary variable for regressor vec rgr; //! data link: H -> rgr datalink rgrlnk; //! model of Y - linear Gaussian mlnorm model; //! options bool opt_L_theta; //! loggers int L_theta; int L_R; int dt_size; public: void getdata ( vec &dt ) { //it_assert_debug ( dt.length() ==Drv.count(),"ArxDS" ); dt=H; }; void getdata ( vec &dt, const ivec &indices ) { it_assert_debug ( dt.length() ==indices.length(),"ArxDS" ); dt=H ( indices ); }; void write ( vec &ut ) { //it_assert_debug ( ut.length() ==Urv.count(),"ArxDS" ); U=ut; }; void write ( vec &ut, const ivec &indices ) { it_assert_debug ( ut.length() ==indices.length(),"ArxDS" ); set_subvector ( U, indices,ut ); }; void step(); //!Default constructor ArxDS ( ) {}; //! Set parameters of the internal model, H is maximum time delay void set_parameters ( const mat &Th0, const vec mu0, const chmat &sqR0 ) { model.set_parameters ( Th0, mu0, sqR0 );}; //! Set void set_drv ( RV &yrv, RV &urv, RV &rrv ) { Rrv = rrv; Urv = urv; dt_size = yrv._dsize() +urv._dsize(); RV drv = concat ( yrv,urv ); Drv = drv; int td = rrv.mint(); H.set_size ( drv._dsize() * ( -td+1 ) ); U.set_size ( Urv._dsize() ); for ( int i=-1;i>=td;i-- ) { drv.t ( -1 ); Drv.add ( drv ); //shift u1 } rgrlnk.set_connection ( rrv,Drv ); dtsize = Drv._dsize(); utsize = Urv._dsize(); } //! set options from a string void set_options ( const string &s ) { opt_L_theta= ( s.find ( "L_theta" ) !=string::npos ); }; virtual void log_add ( logger &L ) { //DS::log_add ( L ); too long!! L_dt=L.add ( Drv ( 0,dt_size ),"" ); L_ut=L.add ( Urv,"" ); mat &A =model._A(); mat R =model._R(); if ( opt_L_theta ) {L_theta=L.add ( RV ( "{th }", vec_1 ( A.rows() *A.cols() ) ),"t" );} if ( opt_L_theta ) {L_R=L.add ( RV ( "{R }", vec_1 ( R.rows() *R.cols() ) ),"r" );} } virtual void logit ( logger &L ) { //DS::logit ( L ); L.logit ( L_dt, H.left ( dt_size ) ); L.logit ( L_ut, U ); mat &A =model._A(); mat R =model._R(); if ( opt_L_theta ) {L.logit ( L_theta,vec ( A._data(), A.rows() *A.cols() ) );}; if ( opt_L_theta ) {L.logit ( L_R, vec ( R._data(), R.rows() *R.rows() ) );}; } }; class stateDS : public DS { protected: //!conditional pdf of the state evolution \f$ f(x_t|x_{t-1}) \f$ mpdf* IM; //!conditional pdf of the observations \f$ f(d_t|x_t) \f$ mpdf* OM; //! result storage vec dt; //! state storage vec xt; //! input storage vec ut; //! Logger int L_xt; public: void getdata ( vec &dt0 ) {dt0=dt;} void getdata ( vec &dt0, const ivec &indeces ) {dt0=dt ( indeces );} stateDS ( mpdf* IM0, mpdf* OM0, int usize ) :DS ( ),IM ( IM0 ),OM ( OM0 ), dt ( OM0->dimension() ), xt ( IM0->dimension() ), ut ( usize ) {} ~stateDS() {delete IM; delete OM;} virtual void step() { xt=IM->samplecond ( concat ( xt,ut ) ); dt=OM->samplecond ( concat ( xt,ut ) ); }; virtual void log_add ( logger &L ) { DS::log_add ( L ); L_xt=L.add ( IM->_rv(),"true" ); } virtual void logit ( logger &L ) { DS::logit ( L ); L.logit ( L_xt,xt ); } }; }; //namespace #endif // DS_H