| 1 | /*! |
|---|
| 2 | \file |
|---|
| 3 | \brief Bayesian Filtering for extensions of autoregressive (ARX) model |
|---|
| 4 | \author Vaclav Smidl. |
|---|
| 5 | |
|---|
| 6 | */ |
|---|
| 7 | |
|---|
| 8 | #ifndef AREX_H |
|---|
| 9 | #define AREX_H |
|---|
| 10 | |
|---|
| 11 | #include "../math/functions.h" |
|---|
| 12 | #include "arx.h" |
|---|
| 13 | |
|---|
| 14 | namespace bdm { |
|---|
| 15 | |
|---|
| 16 | /*! |
|---|
| 17 | * \brief Non-linear transformation + Gaussian noise |
|---|
| 18 | |
|---|
| 19 | Regression of the following kind: |
|---|
| 20 | \f[ |
|---|
| 21 | y_t = g(\psi_t ) + r e_t |
|---|
| 22 | \f] |
|---|
| 23 | where unknown parameters \c rv are \f$[r]\f$, regression vector \f$\psi_t=\psi(y_{1:t},u_{1:t})\f$ is propagated through a known function \c g. Distrubances \f$e_t\f$ are supposed to be normally distributed: |
|---|
| 24 | \f[ |
|---|
| 25 | e_t \sim \mathcal{N}(0,1). |
|---|
| 26 | \f] |
|---|
| 27 | |
|---|
| 28 | */ |
|---|
| 29 | class ARXg: public ARX { |
|---|
| 30 | protected: |
|---|
| 31 | //! function g |
|---|
| 32 | shared_ptr<fnc> g; |
|---|
| 33 | public: |
|---|
| 34 | //! \name Constructors |
|---|
| 35 | //!@{ |
|---|
| 36 | ARXg (): ARX() {} |
|---|
| 37 | ARXg (const ARXg &A0): ARX(A0),g(A0.g) {} |
|---|
| 38 | ARXg* _copy() const { |
|---|
| 39 | return new ARXg(*this); |
|---|
| 40 | } |
|---|
| 41 | //!@} |
|---|
| 42 | |
|---|
| 43 | //!\name Mathematical operations |
|---|
| 44 | //!@{ |
|---|
| 45 | |
|---|
| 46 | //! Weighted Bayes \f$ dt = [y_t psi_t] \f$. |
|---|
| 47 | void bayes_weighted ( const vec &yt, const vec &cond = empty_vec, const double w = 1.0 ) { |
|---|
| 48 | int dimc_store =dimc; |
|---|
| 49 | dimc =0; |
|---|
| 50 | int rgrlen_store = rgrlen; |
|---|
| 51 | rgrlen =0; |
|---|
| 52 | ARX::bayes_weighted(yt - g->eval(cond), empty_vec, w); |
|---|
| 53 | dimc = dimc_store; |
|---|
| 54 | rgrlen=rgrlen_store; |
|---|
| 55 | }; |
|---|
| 56 | //!@} |
|---|
| 57 | |
|---|
| 58 | /*! UI for ARXg estimator |
|---|
| 59 | |
|---|
| 60 | \code |
|---|
| 61 | class = 'ARXg'; |
|---|
| 62 | yrv = RV({names_of_dt} ) // description of output variables |
|---|
| 63 | rgr = RV({names_of_regressors}, [-1,-2]} // description of regressor variables |
|---|
| 64 | constant = 1; // 0/1 switch if the constant term is modelled or not |
|---|
| 65 | g = {class='my_function',...} // function transforming regressor |
|---|
| 66 | |
|---|
| 67 | --- optional --- |
|---|
| 68 | prior = {class='egiw',...}; // Prior density, when given default is used instead |
|---|
| 69 | alternative = {class='egiw',...}; // Alternative density in stabilized estimation, when not given prior is used |
|---|
| 70 | |
|---|
| 71 | frg = 1.0; // forgetting, default frg=1.0 |
|---|
| 72 | |
|---|
| 73 | rv = RV({names_of_parameters}} // description of parametetr names |
|---|
| 74 | \endcode |
|---|
| 75 | */ |
|---|
| 76 | void from_setting ( const Setting &set ) { |
|---|
| 77 | ARX::from_setting(set); |
|---|
| 78 | g=UI::build<fnc>(set,"g",UI::compulsory); |
|---|
| 79 | } |
|---|
| 80 | |
|---|
| 81 | void validate() { |
|---|
| 82 | ARX::validate();//if dimc not set set it from V |
|---|
| 83 | bdm_assert(g->dimension()==dimensiony(),"incompatible g"); |
|---|
| 84 | bdm_assert(g->dimensionc()==dimensionc(),"incompatible g"); |
|---|
| 85 | } |
|---|
| 86 | |
|---|
| 87 | void to_setting ( Setting &set ) const |
|---|
| 88 | { |
|---|
| 89 | ARX::to_setting( set ); // takes care of rv, yrv, rvc |
|---|
| 90 | UI::save(g,set,"g"); |
|---|
| 91 | } |
|---|
| 92 | }; |
|---|
| 93 | UIREGISTER ( ARXg ); |
|---|
| 94 | SHAREDPTR ( ARXg ); |
|---|
| 95 | |
|---|
| 96 | }; |
|---|
| 97 | #endif // AREX_H |
|---|
| 98 | |
|---|
