[18] | 1 | /*! |
---|
| 2 | \file |
---|
| 3 | \brief Common DataSources. |
---|
| 4 | \author Vaclav Smidl. |
---|
| 5 | |
---|
| 6 | ----------------------------------- |
---|
| 7 | BDM++ - C++ library for Bayesian Decision Making under Uncertainty |
---|
| 8 | |
---|
| 9 | Using IT++ for numerical operations |
---|
| 10 | ----------------------------------- |
---|
| 11 | */ |
---|
| 12 | |
---|
[384] | 13 | #ifndef DATASOURCE_H |
---|
| 14 | #define DATASOURCE_H |
---|
[18] | 15 | |
---|
[263] | 16 | |
---|
[384] | 17 | #include "../base/bdmbase.h" |
---|
| 18 | #include "../stat/exp_family.h" |
---|
| 19 | #include "../base/user_info.h" |
---|
[18] | 20 | |
---|
[263] | 21 | namespace bdm { |
---|
[271] | 22 | /*! |
---|
| 23 | * \brief Memory storage of off-line data column-wise |
---|
[18] | 24 | |
---|
[611] | 25 | 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. |
---|
[18] | 26 | |
---|
[271] | 27 | The data can be loaded from a file. |
---|
| 28 | */ |
---|
| 29 | class MemDS : public DS { |
---|
[598] | 30 | protected: |
---|
| 31 | //! internal matrix of data |
---|
| 32 | mat Data; |
---|
| 33 | //! active column in the Data matrix |
---|
| 34 | int time; |
---|
| 35 | //! vector of rows that are presented in Dt |
---|
| 36 | ivec rowid; |
---|
[18] | 37 | |
---|
[598] | 38 | public: |
---|
[609] | 39 | int max_length() {return Data.cols();} |
---|
[676] | 40 | void getdata ( vec &dt ) const; |
---|
[598] | 41 | void getdata ( vec &dt, const ivec &indeces ); |
---|
[660] | 42 | void set_drv (const RV &drv,const RV &urv ); |
---|
[565] | 43 | |
---|
[660] | 44 | void write ( const vec &ut ) { |
---|
[598] | 45 | bdm_error ( "MemDS::write is not supported" ); |
---|
| 46 | } |
---|
[565] | 47 | |
---|
[660] | 48 | void write ( const vec &ut, const ivec &indices ) { |
---|
[598] | 49 | bdm_error ( "MemDS::write is not supported" ); |
---|
| 50 | } |
---|
[565] | 51 | |
---|
[598] | 52 | void step(); |
---|
| 53 | //!Default constructor |
---|
| 54 | MemDS () {}; |
---|
[660] | 55 | //! Convenience constructor |
---|
[609] | 56 | MemDS ( mat &Dat, ivec &rowid0); |
---|
| 57 | /*! Create object from the following structure |
---|
| 58 | \code |
---|
[611] | 59 | { class = "MemDS"; |
---|
[609] | 60 | Data = (...); // Data matrix or data vector |
---|
| 61 | --- optional --- |
---|
[611] | 62 | drv = {class="RV"; ...} // Identification how rows of the matrix Data will be known to others |
---|
[609] | 63 | time = 0; // Index of the first column to user_info, |
---|
| 64 | rowid = [1,2,3...]; // ids of rows to be used |
---|
| 65 | } |
---|
| 66 | \endcode |
---|
| 67 | |
---|
| 68 | If the optional fields are not given, they will be filled as follows: |
---|
| 69 | \code |
---|
| 70 | rowid= [0, 1, 2, ...number_of_rows_of_Data]; |
---|
[611] | 71 | drv = {names=("ch0", "ch1", "ch2", ..."number_of_rows_of_Data"); |
---|
[609] | 72 | sizes=( 1 1 1 ...); |
---|
| 73 | times=( 0 0 0 ...); |
---|
| 74 | }; |
---|
| 75 | time = 0; |
---|
| 76 | \endcode |
---|
| 77 | If \c rowid is given, \c drv will be named after indeces in rowids. |
---|
| 78 | |
---|
| 79 | Hence the data provided by method \c getdata() will be full column of matrix Data starting from the first record. |
---|
| 80 | */ |
---|
| 81 | void from_setting(const Setting &set){ |
---|
| 82 | UI::get(Data, set, "Data", UI::compulsory); |
---|
| 83 | if(!UI::get(time, set,"time", UI::optional)) {time =0;} |
---|
| 84 | if(!UI::get(rowid, set, "rowid",UI::optional)) {rowid =linspace(0,Data.rows()-1);} |
---|
| 85 | shared_ptr<RV> r=UI::build<RV>(set,"drv",UI::optional); |
---|
| 86 | if (!r) {r=new RV(); |
---|
| 87 | for (int i=0; i<rowid.length(); i++){ r->add(RV("ch"+num2str(rowid(i)), 1, 0));} |
---|
| 88 | } |
---|
| 89 | set_drv(*r,RV()); //empty urv |
---|
| 90 | dtsize=r->_dsize(); |
---|
[676] | 91 | ytsize = dtsize; |
---|
[609] | 92 | utsize=0; |
---|
| 93 | } |
---|
[271] | 94 | }; |
---|
[609] | 95 | UIREGISTER(MemDS); |
---|
[263] | 96 | |
---|
[611] | 97 | /*! \brief Simulate data from a static pdf (epdf) |
---|
| 98 | |
---|
[598] | 99 | Trivial example of a data source, could be used for tests of some estimation algorithms. For example, simulating data from a mixture model and feeding them to mixture model estimators. |
---|
| 100 | */ |
---|
| 101 | |
---|
| 102 | class EpdfDS: public DS { |
---|
| 103 | protected: |
---|
| 104 | //! internal pointer to epdf from which we samplecond |
---|
| 105 | shared_ptr<epdf> iepdf; |
---|
| 106 | //! internal storage of data sample |
---|
| 107 | vec dt; |
---|
| 108 | public: |
---|
| 109 | void step() { |
---|
| 110 | dt=iepdf->sample(); |
---|
| 111 | } |
---|
[676] | 112 | void getdata ( vec &dt_out ) const { |
---|
[598] | 113 | dt_out = dt; |
---|
| 114 | } |
---|
| 115 | void getdata ( vec &dt_out, const ivec &ids ) { |
---|
| 116 | dt_out = dt ( ids ); |
---|
| 117 | } |
---|
[660] | 118 | const RV& _drv() const { |
---|
[598] | 119 | return iepdf->_rv(); |
---|
| 120 | } |
---|
| 121 | |
---|
| 122 | /*! |
---|
| 123 | \code |
---|
[611] | 124 | class = "EpdfDS"; |
---|
| 125 | epdf = {class="epdf_offspring", ...}// uncondtitional density to sample from |
---|
[598] | 126 | \endcode |
---|
| 127 | |
---|
| 128 | */ |
---|
| 129 | void from_setting ( const Setting &set ) { |
---|
| 130 | iepdf=UI::build<epdf> ( set,"epdf",UI::compulsory ); |
---|
[611] | 131 | bdm_assert(iepdf->isnamed(), "Input epdf must be named, check if RV is given correctly"); |
---|
[613] | 132 | dt = zeros(iepdf->dimension()); |
---|
[611] | 133 | dtsize=dt.length(); |
---|
[700] | 134 | ytsize=dt.length(); |
---|
[611] | 135 | set_drv(iepdf->_rv(),RV()); |
---|
| 136 | utsize =0; |
---|
[618] | 137 | validate(); |
---|
[598] | 138 | } |
---|
[613] | 139 | void validate() { |
---|
| 140 | dt = iepdf->sample(); |
---|
| 141 | } |
---|
[598] | 142 | }; |
---|
| 143 | UIREGISTER ( EpdfDS ); |
---|
| 144 | |
---|
| 145 | /*! \brief Simulate data from conditional density |
---|
| 146 | Still having only one density but allowing conditioning on either input or delayed values. |
---|
| 147 | */ |
---|
[693] | 148 | class PdfDS :public DS { |
---|
[598] | 149 | protected: |
---|
| 150 | //! internal pointer to epdf from which we samplecond |
---|
[693] | 151 | shared_ptr<pdf> ipdf; |
---|
[598] | 152 | //! internal storage of data sample |
---|
[603] | 153 | vec yt; |
---|
[598] | 154 | //! input vector |
---|
| 155 | vec ut; |
---|
[603] | 156 | //! datalink between ut and regressor |
---|
| 157 | datalink_buffered ut2rgr; |
---|
| 158 | //! datalink between yt and regressor |
---|
| 159 | datalink_buffered yt2rgr; |
---|
[598] | 160 | //! numeric values of regressor |
---|
| 161 | vec rgr; |
---|
| 162 | |
---|
| 163 | public: |
---|
| 164 | void step() { |
---|
[603] | 165 | yt2rgr.step(yt); // y is now history |
---|
| 166 | ut2rgr.filldown ( ut,rgr ); |
---|
| 167 | yt2rgr.filldown ( yt,rgr ); |
---|
[693] | 168 | yt=ipdf->samplecond ( rgr ); |
---|
[603] | 169 | ut2rgr.step(ut); //u is now history |
---|
[598] | 170 | } |
---|
[676] | 171 | void getdata ( vec &dt_out ) const { |
---|
[603] | 172 | bdm_assert_debug(dt_out.length()>=utsize+ytsize,"Short output vector"); |
---|
| 173 | dt_out.set_subvector(0, yt); |
---|
| 174 | dt_out.set_subvector(ytsize, ut); |
---|
[598] | 175 | } |
---|
| 176 | void write(const vec &ut0){ut=ut0;} |
---|
| 177 | |
---|
| 178 | /*! |
---|
| 179 | \code |
---|
[693] | 180 | class = "PdfDS"; |
---|
| 181 | pdf = {class="pdf_offspring", ...}; // pdf to simulate |
---|
[613] | 182 | --- optional --- |
---|
| 183 | init_rv = {class="RV",names=...}; // define what rv to initialize - typically delayed values! |
---|
| 184 | init_values = [...]; // vector of initial values corresponding to init_rv |
---|
[598] | 185 | \endcode |
---|
| 186 | |
---|
[613] | 187 | If init_rv is not given, init_values are set to zero. |
---|
[598] | 188 | */ |
---|
| 189 | void from_setting ( const Setting &set ) { |
---|
[693] | 190 | ipdf=UI::build<pdf> ( set,"pdf",UI::compulsory ); |
---|
[598] | 191 | |
---|
[693] | 192 | Yrv = ipdf->_rv(); |
---|
[598] | 193 | // get unique rvs form rvc |
---|
[693] | 194 | RV rgrv0=ipdf->_rvc().remove_time(); |
---|
[603] | 195 | // input is what in not in Yrv |
---|
| 196 | Urv=rgrv0.subt(Yrv); |
---|
| 197 | set_drv(Yrv, Urv); |
---|
[598] | 198 | // connect input and output to rvc |
---|
[693] | 199 | ut2rgr.set_connection(ipdf->_rvc(), Urv); |
---|
| 200 | yt2rgr.set_connection(ipdf->_rvc(), Yrv); |
---|
[613] | 201 | |
---|
| 202 | //set history - if given |
---|
| 203 | shared_ptr<RV> rv_ini=UI::build<RV>(set,"init_rv",UI::optional); |
---|
| 204 | if(rv_ini){ // check if |
---|
| 205 | vec val; |
---|
| 206 | UI::get(val, set, "init_values", UI::optional); |
---|
| 207 | if (val.length()!=rv_ini->_dsize()){ |
---|
| 208 | bdm_error("init_rv and init_values fields have incompatible sizes"); |
---|
| 209 | } else { |
---|
| 210 | ut2rgr.set_history(*rv_ini, val); |
---|
| 211 | yt2rgr.set_history(*rv_ini, val); |
---|
| 212 | } |
---|
| 213 | } |
---|
[598] | 214 | |
---|
[693] | 215 | yt = zeros ( ipdf->dimension() ); |
---|
| 216 | rgr = zeros ( ipdf->dimensionc() ); |
---|
[603] | 217 | ut = zeros(Urv._dsize()); |
---|
| 218 | |
---|
| 219 | ytsize=yt.length(); |
---|
| 220 | utsize=ut.length(); |
---|
| 221 | dtsize = ytsize+utsize; |
---|
[613] | 222 | validate(); |
---|
[598] | 223 | } |
---|
[613] | 224 | void validate() { |
---|
| 225 | //taken from sample() - shift of history is not done here |
---|
| 226 | ut2rgr.filldown ( ut,rgr ); |
---|
| 227 | yt2rgr.filldown ( yt,rgr ); |
---|
[693] | 228 | yt=ipdf->samplecond ( rgr ); |
---|
[613] | 229 | } |
---|
[598] | 230 | }; |
---|
[693] | 231 | UIREGISTER ( PdfDS ); |
---|
[598] | 232 | |
---|
[308] | 233 | /*! Pseudovirtual class for reading data from files |
---|
[283] | 234 | |
---|
| 235 | */ |
---|
| 236 | class FileDS: public MemDS { |
---|
| 237 | |
---|
[598] | 238 | public: |
---|
| 239 | void getdata ( vec &dt ) { |
---|
| 240 | dt = Data.get_col ( time ); |
---|
| 241 | } |
---|
[565] | 242 | |
---|
[598] | 243 | void getdata ( vec &dt, const ivec &indices ) { |
---|
| 244 | vec tmp = Data.get_col ( time ); |
---|
| 245 | dt = tmp ( indices ); |
---|
| 246 | } |
---|
[565] | 247 | |
---|
[598] | 248 | //! returns number of data in the file; |
---|
| 249 | int ndat() { |
---|
| 250 | return Data.cols(); |
---|
| 251 | } |
---|
| 252 | //! no sense to log this type |
---|
[676] | 253 | void log_register(logger &L, const string &prefix){}; |
---|
[598] | 254 | //! no sense to log this type |
---|
[676] | 255 | void log_write ( ) const {}; |
---|
[283] | 256 | }; |
---|
| 257 | |
---|
[477] | 258 | /*! |
---|
[308] | 259 | * \brief Read Data Matrix from an IT file |
---|
| 260 | |
---|
| 261 | 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$t\f$! |
---|
| 262 | |
---|
| 263 | */ |
---|
[342] | 264 | class ITppFileDS: public FileDS { |
---|
[308] | 265 | |
---|
[598] | 266 | public: |
---|
[660] | 267 | //! Convenience constructor |
---|
[598] | 268 | ITppFileDS ( const string &fname, const string &varname ) : FileDS() { |
---|
| 269 | it_file it ( fname ); |
---|
| 270 | it << Name ( varname ); |
---|
| 271 | it >> Data; |
---|
| 272 | time = 0; |
---|
| 273 | //rowid and delays are ignored |
---|
| 274 | }; |
---|
[357] | 275 | |
---|
[598] | 276 | ITppFileDS () : FileDS() { |
---|
| 277 | }; |
---|
[357] | 278 | |
---|
[598] | 279 | void from_setting ( const Setting &set ); |
---|
[357] | 280 | |
---|
[598] | 281 | // TODO dodelat void to_setting( Setting &set ) const; |
---|
[357] | 282 | |
---|
[308] | 283 | }; |
---|
| 284 | |
---|
[477] | 285 | UIREGISTER ( ITppFileDS ); |
---|
[529] | 286 | SHAREDPTR ( ITppFileDS ); |
---|
[357] | 287 | |
---|
[271] | 288 | /*! |
---|
[308] | 289 | * \brief CSV file data storage |
---|
| 290 | The constructor creates \c Data matrix from the records in a CSV file \c fname. The orientation can be of two types: |
---|
| 291 | 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. |
---|
| 292 | 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. |
---|
| 293 | |
---|
| 294 | */ |
---|
| 295 | class CsvFileDS: public FileDS { |
---|
| 296 | |
---|
[598] | 297 | public: |
---|
| 298 | //! Constructor - create DS from a CSV file. |
---|
| 299 | CsvFileDS ( const string& fname, const string& orientation = "BY_COL" ); |
---|
[308] | 300 | }; |
---|
| 301 | |
---|
| 302 | |
---|
| 303 | |
---|
[660] | 304 | // ARXDs - DELETED |
---|
[263] | 305 | |
---|
[660] | 306 | //! State-space data source simulating two densities |
---|
[695] | 307 | class StateDS : public DS { |
---|
| 308 | protected: |
---|
[598] | 309 | //!conditional pdf of the state evolution \f$ f(x_t|x_{t-1}) \f$ |
---|
[697] | 310 | shared_ptr<pdf> IM; |
---|
[527] | 311 | |
---|
[598] | 312 | //!conditional pdf of the observations \f$ f(d_t|x_t) \f$ |
---|
[697] | 313 | shared_ptr<pdf> OM; |
---|
[527] | 314 | |
---|
[598] | 315 | //! result storage |
---|
| 316 | vec dt; |
---|
| 317 | //! state storage |
---|
| 318 | vec xt; |
---|
| 319 | //! input storage |
---|
| 320 | vec ut; |
---|
[527] | 321 | |
---|
[695] | 322 | //! datalink from ut to IM.rvc |
---|
| 323 | datalink_part u2imc; |
---|
| 324 | //! datalink from ut to OM.rvc |
---|
| 325 | datalink_part u2omc; |
---|
[598] | 326 | public: |
---|
[695] | 327 | void getdata ( vec &dt0 ) const { |
---|
[598] | 328 | dt0 = dt; |
---|
| 329 | } |
---|
[695] | 330 | void write (const vec &ut0 ) { |
---|
| 331 | ut = ut0; |
---|
| 332 | } |
---|
| 333 | |
---|
[598] | 334 | void getdata ( vec &dt0, const ivec &indices ) { |
---|
| 335 | dt0 = dt ( indices ); |
---|
| 336 | } |
---|
[357] | 337 | |
---|
[598] | 338 | virtual void step() { |
---|
[695] | 339 | vec imc(IM->dimensionc()); |
---|
| 340 | imc.set_subvector(0,xt); |
---|
| 341 | u2imc.filldown(ut,imc); |
---|
| 342 | xt = IM->samplecond ( imc ); |
---|
| 343 | |
---|
| 344 | vec omc(OM->dimensionc()); |
---|
| 345 | omc.set_subvector(0,xt); |
---|
| 346 | u2omc.filldown(ut,omc); |
---|
| 347 | vec yt; |
---|
| 348 | yt = OM->samplecond ( omc ); |
---|
| 349 | //fill all data |
---|
| 350 | dt.set_subvector(0,yt); |
---|
| 351 | dt.set_subvector(yt.length(),xt); |
---|
| 352 | dt.set_subvector(ytsize,ut); |
---|
[598] | 353 | } |
---|
[267] | 354 | |
---|
[695] | 355 | //! set parameters |
---|
[697] | 356 | void set_parameters(shared_ptr<pdf> IM0, shared_ptr<pdf> OM0){ |
---|
[695] | 357 | IM=IM0; |
---|
| 358 | OM = OM0; |
---|
[598] | 359 | } |
---|
[695] | 360 | void set_initx(const vec &x0){xt=x0;} |
---|
[267] | 361 | |
---|
[598] | 362 | /*! UI for stateDS |
---|
[357] | 363 | |
---|
[598] | 364 | The DS is constructed from a structure with fields: |
---|
| 365 | \code |
---|
[695] | 366 | class = "stateDS"; |
---|
| 367 | //Internal model |
---|
[700] | 368 | IM = { type = "pdf-offspring"; }; |
---|
[695] | 369 | //Observation model |
---|
[700] | 370 | OM = { type = "pdf-offspring"; } |
---|
[695] | 371 | //initial state |
---|
| 372 | x0 = [...]; //vector of data |
---|
[598] | 373 | \endcode |
---|
[695] | 374 | Both models must have defined \c rv. and \c rvc |
---|
| 375 | Random variables not found in any rv are considered to be inputs. |
---|
[598] | 376 | */ |
---|
| 377 | void from_setting ( const Setting &set ); |
---|
[357] | 378 | |
---|
[598] | 379 | // TODO dodelat void to_setting( Setting &set ) const; |
---|
[357] | 380 | |
---|
[695] | 381 | void validate(); |
---|
[271] | 382 | }; |
---|
[267] | 383 | |
---|
[695] | 384 | UIREGISTER ( StateDS ); |
---|
| 385 | SHAREDPTR ( StateDS ); |
---|
[357] | 386 | |
---|
[254] | 387 | }; //namespace |
---|
[18] | 388 | |
---|
| 389 | #endif // DS_H |
---|