root/doc/html/libKF_8h-source.html @ 32

Revision 32, 29.5 kB (checked in by smidl, 17 years ago)

test KF : estimation of R in KF is not possible! Likelihood of y_t is growing when R -> 0

RevLine 
[8]1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
2<html><head><meta http-equiv="Content-Type" content="text/html;charset=UTF-8">
[19]3<title>mixpp: work/mixpp/bdm/estim/libKF.h Source File</title>
[8]4<link href="doxygen.css" rel="stylesheet" type="text/css">
5<link href="tabs.css" rel="stylesheet" type="text/css">
6</head><body>
7<!-- Generated by Doxygen 1.5.3 -->
8<div class="tabs">
9  <ul>
10    <li><a href="index.html"><span>Main&nbsp;Page</span></a></li>
11    <li><a href="annotated.html"><span>Classes</span></a></li>
12    <li class="current"><a href="files.html"><span>Files</span></a></li>
[28]13    <li><a href="pages.html"><span>Related&nbsp;Pages</span></a></li>
[8]14  </ul>
15</div>
[19]16<h1>work/mixpp/bdm/estim/libKF.h</h1><a href="libKF_8h.html">Go to the documentation of this file.</a><div class="fragment"><pre class="fragment"><a name="l00001"></a>00001
[8]17<a name="l00013"></a>00013 <span class="preprocessor">#ifndef KF_H</span>
18<a name="l00014"></a>00014 <span class="preprocessor"></span><span class="preprocessor">#define KF_H</span>
19<a name="l00015"></a>00015 <span class="preprocessor"></span>
20<a name="l00016"></a>00016 <span class="preprocessor">#include &lt;itpp/itbase.h&gt;</span>
[22]21<a name="l00017"></a>00017 <span class="preprocessor">#include "../stat/libFN.h"</span>
[32]22<a name="l00018"></a>00018 <span class="preprocessor">#include "../stat/libEF.h"</span>
[8]23<a name="l00019"></a>00019
24<a name="l00020"></a>00020
25<a name="l00021"></a>00021 <span class="keyword">using namespace </span>itpp;
26<a name="l00022"></a>00022
[32]27<a name="l00026"></a><a class="code" href="classKalmanFull.html">00026</a> <span class="keyword">class </span><a class="code" href="classKalmanFull.html" title="Basic Kalman filter with full matrices (education purpose only)! Will be deleted...">KalmanFull</a> {
[8]28<a name="l00027"></a>00027         <span class="keywordtype">int</span> dimx, dimy, dimu;
29<a name="l00028"></a>00028         mat A, B, C, D, R, Q;
30<a name="l00029"></a>00029         
31<a name="l00030"></a>00030         <span class="comment">//cache </span>
32<a name="l00031"></a>00031         mat _Pp, _Ry, _iRy, _K;
33<a name="l00032"></a>00032 <span class="keyword">public</span>:
34<a name="l00033"></a>00033         <span class="comment">//posterior </span>
35<a name="l00035"></a><a class="code" href="classKalmanFull.html#fb5aec635e2720cc5ac31bc01c18a68a">00035</a> <span class="comment"></span>        vec <a class="code" href="classKalmanFull.html#fb5aec635e2720cc5ac31bc01c18a68a" title="Mean value of the posterior density.">mu</a>;
36<a name="l00037"></a><a class="code" href="classKalmanFull.html#b75dc059e84fa8ffc076203b30f926cc">00037</a>         mat <a class="code" href="classKalmanFull.html#b75dc059e84fa8ffc076203b30f926cc" title="Variance of the posterior density.">P</a>;
37<a name="l00038"></a>00038
38<a name="l00039"></a>00039 <span class="keyword">public</span>:
39<a name="l00041"></a>00041         <a class="code" href="classKalmanFull.html#7197ab6e7380790006394eabd3b97043" title="Full constructor.">KalmanFull</a> ( mat A, mat B, mat C, mat D, mat R, mat Q, mat P0, vec mu0);
[32]40<a name="l00043"></a>00043         <span class="keywordtype">void</span> <a class="code" href="classKalmanFull.html#13a041cd98ff157703766be275a657bb" title="Here dt = [yt;ut] of appropriate dimensions.">bayes</a>(<span class="keyword">const</span> vec &amp;dt);
[8]41<a name="l00044"></a>00044
42<a name="l00045"></a>00045         <span class="keyword">friend</span> std::ostream &amp;operator&lt;&lt; ( std::ostream &amp;os, <span class="keyword">const</span> <a class="code" href="classKalmanFull.html" title="Basic Kalman filter with full matrices (education purpose only)! Will be deleted...">KalmanFull</a> &amp;kf );
43<a name="l00046"></a>00046
44<a name="l00047"></a>00047 };
45<a name="l00048"></a>00048
46<a name="l00049"></a>00049
47<a name="l00053"></a>00053 <span class="keyword">template</span>&lt;<span class="keyword">class</span> sq_T&gt;
[22]48<a name="l00054"></a><a class="code" href="classKalman.html">00054</a> <span class="keyword">class </span><a class="code" href="classKalman.html" title="Kalman filter with covariance matrices in square root form.">Kalman</a> : <span class="keyword">public</span> <a class="code" href="classBM.html" title="Bayesian Model of the world, i.e. all uncertainty is modeled by probabilities.">BM</a> {
49<a name="l00055"></a>00055 <span class="keyword">protected</span>:
[32]50<a name="l00056"></a>00056         <a class="code" href="classRV.html" title="Class representing variables, most often random variables.">RV</a> rvy;
51<a name="l00057"></a>00057         <a class="code" href="classRV.html" title="Class representing variables, most often random variables.">RV</a> rvu;
52<a name="l00058"></a>00058         <span class="keywordtype">int</span> dimx, dimy, dimu;
53<a name="l00059"></a>00059         mat A, B, C, D;
54<a name="l00060"></a>00060         sq_T R, Q;
55<a name="l00061"></a>00061         
56<a name="l00063"></a><a class="code" href="classKalman.html#5568c74bac67ae6d3b1061dba60c9424">00063</a>         <a class="code" href="classenorm.html" title="Gaussian density with positive definite (decomposed) covariance matrix.">enorm&lt;sq_T&gt;</a> <a class="code" href="classKalman.html#5568c74bac67ae6d3b1061dba60c9424" title="posterior density on $x_t$">est</a>;
57<a name="l00065"></a><a class="code" href="classKalman.html#e580ab06483952bd03f2e651763e184f">00065</a>         <a class="code" href="classenorm.html" title="Gaussian density with positive definite (decomposed) covariance matrix.">enorm&lt;sq_T&gt;</a> <a class="code" href="classKalman.html#e580ab06483952bd03f2e651763e184f" title="preditive density on $y_t$">fy</a>;
58<a name="l00066"></a>00066         
59<a name="l00067"></a>00067         mat _K;
60<a name="l00068"></a>00068         <span class="comment">//cache of fy</span>
61<a name="l00069"></a>00069         vec* _yp;
62<a name="l00070"></a>00070         sq_T* _Ry,*_iRy;
63<a name="l00071"></a>00071         <span class="comment">//cache of est</span>
64<a name="l00072"></a>00072         vec* _mu;
65<a name="l00073"></a>00073         sq_T* _P, *_iP;
66<a name="l00074"></a>00074         
67<a name="l00075"></a>00075 <span class="keyword">public</span>:
68<a name="l00077"></a>00077         <a class="code" href="classKalman.html#3d56b0a97b8c1e25fdd3b10eef3c2ad3" title="Default constructor.">Kalman</a> (<a class="code" href="classRV.html" title="Class representing variables, most often random variables.">RV</a> rvx0, <a class="code" href="classRV.html" title="Class representing variables, most often random variables.">RV</a> rvy0, <a class="code" href="classRV.html" title="Class representing variables, most often random variables.">RV</a> rvu0);
69<a name="l00079"></a>00079         <a class="code" href="classKalman.html#3d56b0a97b8c1e25fdd3b10eef3c2ad3" title="Default constructor.">Kalman</a> (<span class="keyword">const</span> <a class="code" href="classKalman.html" title="Kalman filter with covariance matrices in square root form.">Kalman&lt;sq_T&gt;</a> &amp;K0);
70<a name="l00081"></a>00081         <span class="keywordtype">void</span> <a class="code" href="classKalman.html#239b28a0380946f5749b2f8d2807f93a" title="Set parameters with check of relevance.">set_parameters</a> (<span class="keyword">const</span> mat &amp;A0,<span class="keyword">const</span> mat &amp;B0,<span class="keyword">const</span> mat &amp;C0,<span class="keyword">const</span> mat &amp;D0,<span class="keyword">const</span> sq_T &amp;R0,<span class="keyword">const</span> sq_T &amp;Q0);
71<a name="l00083"></a><a class="code" href="classKalman.html#80bcf29466d9a9dd2b8f74699807d0c0">00083</a>         <span class="keywordtype">void</span> <a class="code" href="classKalman.html#80bcf29466d9a9dd2b8f74699807d0c0" title="Set estimate values, used e.g. in initialization.">set_est</a>(<span class="keyword">const</span> vec &amp;mu0, <span class="keyword">const</span> sq_T &amp;P0 ){<a class="code" href="classKalman.html#5568c74bac67ae6d3b1061dba60c9424" title="posterior density on $x_t$">est</a>.set_parameters(mu0,P0);};
72<a name="l00085"></a>00085         <span class="keywordtype">void</span> <a class="code" href="classKalman.html#7750ffd73f261828a32c18aaeb65c75c" title="Here dt = [yt;ut] of appropriate dimensions.">bayes</a>(<span class="keyword">const</span> vec &amp;dt);
73<a name="l00086"></a><a class="code" href="classKalman.html#a213c57aef55b2645e550bed81cfc0d4">00086</a>         <a class="code" href="classepdf.html" title="Probability density function with numerical statistics, e.g. posterior density.">epdf</a>&amp; <a class="code" href="classKalman.html#a213c57aef55b2645e550bed81cfc0d4" title="Returns a pointer to the epdf representing posterior density on parameters. Use with...">_epdf</a>(){<span class="keywordflow">return</span> <a class="code" href="classKalman.html#5568c74bac67ae6d3b1061dba60c9424" title="posterior density on $x_t$">est</a>;}
74<a name="l00087"></a>00087 <span class="comment">//      friend std::ostream &amp;operator&lt;&lt; ( std::ostream &amp;os, const Kalman&lt;sq_T&gt; &amp;kf );</span>
75<a name="l00088"></a>00088
76<a name="l00090"></a>00090 };
77<a name="l00091"></a>00091
78<a name="l00097"></a>00097 <span class="keyword">template</span>&lt;<span class="keyword">class</span> sq_T&gt;
79<a name="l00098"></a><a class="code" href="classEKF.html">00098</a> <span class="keyword">class </span><a class="code" href="classEKF.html" title="Extended Kalman Filter.">EKF</a> : <span class="keyword">public</span> <a class="code" href="classKalman.html" title="Kalman filter with covariance matrices in square root form.">Kalman</a>&lt;ldmat&gt; {
80<a name="l00100"></a>00100         <a class="code" href="classdiffbifn.html" title="Class representing a differentiable function of two variables $f(x,u)$.">diffbifn</a>* pfxu;
81<a name="l00102"></a>00102         <a class="code" href="classdiffbifn.html" title="Class representing a differentiable function of two variables $f(x,u)$.">diffbifn</a>* phxu;
82<a name="l00103"></a>00103 <span class="keyword">public</span>:
83<a name="l00105"></a>00105         <a class="code" href="classEKF.html#ea4f3254cacf0a92d2a820b1201d049e" title="Default constructor.">EKF</a> (<a class="code" href="classRV.html" title="Class representing variables, most often random variables.">RV</a> rvx, <a class="code" href="classRV.html" title="Class representing variables, most often random variables.">RV</a> rvy, <a class="code" href="classRV.html" title="Class representing variables, most often random variables.">RV</a> rvu);
84<a name="l00106"></a>00106         <span class="keywordtype">void</span> set_parameters(<a class="code" href="classdiffbifn.html" title="Class representing a differentiable function of two variables $f(x,u)$.">diffbifn</a>* pfxu, <a class="code" href="classdiffbifn.html" title="Class representing a differentiable function of two variables $f(x,u)$.">diffbifn</a>* phxu, <span class="keyword">const</span> sq_T Q0, <span class="keyword">const</span> sq_T R0);
85<a name="l00108"></a>00108         <span class="keywordtype">void</span> <a class="code" href="classEKF.html#c79c62c9b3e0b56b3aaa1b6f1d9a7af7" title="Here dt = [yt;ut] of appropriate dimensions.">bayes</a>(<span class="keyword">const</span> vec &amp;dt);     
86<a name="l00109"></a>00109 };
87<a name="l00110"></a>00110
88<a name="l00114"></a><a class="code" href="classKFcondQR.html">00114</a> <span class="keyword">class </span><a class="code" href="classKFcondQR.html" title="Kalman Filter with conditional diagonal matrices R and Q.">KFcondQR</a> : <span class="keyword">public</span> <a class="code" href="classKalman.html" title="Kalman filter with covariance matrices in square root form.">Kalman</a>&lt;ldmat&gt;, <span class="keyword">public</span> <a class="code" href="classBMcond.html" title="Conditional Bayesian Filter.">BMcond</a> {
89<a name="l00115"></a>00115 <span class="comment">//protected:</span>
90<a name="l00116"></a>00116 <span class="keyword">public</span>:
91<a name="l00117"></a>00117 <a class="code" href="classKFcondQR.html" title="Kalman Filter with conditional diagonal matrices R and Q.">KFcondQR</a>(<a class="code" href="classRV.html" title="Class representing variables, most often random variables.">RV</a> rvx, <a class="code" href="classRV.html" title="Class representing variables, most often random variables.">RV</a> rvy, <a class="code" href="classRV.html" title="Class representing variables, most often random variables.">RV</a> rvu): <a class="code" href="classKalman.html" title="Kalman filter with covariance matrices in square root form.">Kalman&lt;ldmat&gt;</a>(rvx, rvy,rvu){};
92<a name="l00118"></a>00118 <span class="keywordtype">void</span> <a class="code" href="classKFcondQR.html#c9ecf292a85327aa6309c9fd70ceb606" title="Substitute val for rvc.">condition</a>(<span class="keyword">const</span> vec &amp;RQ);
93<a name="l00119"></a>00119 };
94<a name="l00120"></a>00120
95<a name="l00122"></a>00122
96<a name="l00123"></a>00123 <span class="keyword">template</span>&lt;<span class="keyword">class</span> sq_T&gt;
97<a name="l00124"></a><a class="code" href="classKalman.html#ce38e31810aea4db45a83ad05eaba009">00124</a> <a class="code" href="classKalman.html#3d56b0a97b8c1e25fdd3b10eef3c2ad3" title="Default constructor.">Kalman&lt;sq_T&gt;::Kalman</a>(<span class="keyword">const</span> <a class="code" href="classKalman.html" title="Kalman filter with covariance matrices in square root form.">Kalman&lt;sq_T&gt;</a> &amp;K0): <a class="code" href="classBM.html" title="Bayesian Model of the world, i.e. all uncertainty is modeled by probabilities.">BM</a>(K0.rv),rvy(K0.rvy),rvu(K0.rvu),
98<a name="l00125"></a>00125 dimx(rv.count()), dimy(rvy.count()),dimu(rvu.count()),
99<a name="l00126"></a>00126  A(dimx,dimx), B(dimx,dimu), C(dimy,dimx), D(dimy,dimu),est(rv), fy(rvy){
100<a name="l00127"></a>00127 this-&gt;<a class="code" href="classKalman.html#239b28a0380946f5749b2f8d2807f93a" title="Set parameters with check of relevance.">set_parameters</a>(K0.<a class="code" href="classKalman.html#5e02efe86ee91e9c74b93b425fe060b9">A</a>, K0.<a class="code" href="classKalman.html#dc87704284a6c0bca13bf51f4345a50a">B</a>, K0.<a class="code" href="classKalman.html#86a805cd6515872d1132ad0d6eb5dc13">C</a>, K0.<a class="code" href="classKalman.html#d69f774ba3335c970c1c5b1d182f4dd1">D</a>, K0.<a class="code" href="classKalman.html#11d171dc0e0ab111c56a70f98b97b3ec">R</a>, K0.<a class="code" href="classKalman.html#9b69015c800eb93f3ee49da23a6f55d9">Q</a>);
101<a name="l00128"></a>00128 <span class="comment">//reset copy values in pointers</span>
102<a name="l00129"></a>00129 *_mu = *K0.<a class="code" href="classKalman.html#d1f669b5b3421a070cc75d77b55ba734">_mu</a>;
103<a name="l00130"></a>00130 *_P = *K0.<a class="code" href="classKalman.html#b3388218567128a797e69b109138271d">_P</a>;
104<a name="l00131"></a>00131 *_iP = *K0.<a class="code" href="classKalman.html#b8bb7f870d69993493ba67ce40e7c3e9">_iP</a>;
105<a name="l00132"></a>00132 *_yp = *K0.<a class="code" href="classKalman.html#5188eb0329f8561f0b357af329769bf8">_yp</a>;
106<a name="l00133"></a>00133 *_iRy = *K0.<a class="code" href="classKalman.html#fbbdf31365f5a5674099599200ea193b">_iRy</a>;
107<a name="l00134"></a>00134 *_Ry = *K0.<a class="code" href="classKalman.html#e17dd745daa8a958035a334a56fa4674">_Ry</a>;
[22]108<a name="l00135"></a>00135
[32]109<a name="l00136"></a>00136 }
110<a name="l00137"></a>00137
111<a name="l00138"></a>00138 <span class="keyword">template</span>&lt;<span class="keyword">class</span> sq_T&gt;
112<a name="l00139"></a><a class="code" href="classKalman.html#3d56b0a97b8c1e25fdd3b10eef3c2ad3">00139</a> <a class="code" href="classKalman.html#3d56b0a97b8c1e25fdd3b10eef3c2ad3" title="Default constructor.">Kalman&lt;sq_T&gt;::Kalman</a>(<a class="code" href="classRV.html" title="Class representing variables, most often random variables.">RV</a> rvx, <a class="code" href="classRV.html" title="Class representing variables, most often random variables.">RV</a> rvy0, <a class="code" href="classRV.html" title="Class representing variables, most often random variables.">RV</a> rvu0): <a class="code" href="classBM.html" title="Bayesian Model of the world, i.e. all uncertainty is modeled by probabilities.">BM</a>(rvx),rvy(rvy0),rvu(rvu0),
113<a name="l00140"></a>00140 dimx(rvx.count()), dimy(rvy.count()),dimu(rvu.count()),
114<a name="l00141"></a>00141  A(dimx,dimx), B(dimx,dimu), C(dimy,dimx), D(dimy,dimu),<a class="code" href="classKalman.html#5568c74bac67ae6d3b1061dba60c9424" title="posterior density on $x_t$">est</a>(rvx), <a class="code" href="classKalman.html#e580ab06483952bd03f2e651763e184f" title="preditive density on $y_t$">fy</a>(rvy){
115<a name="l00142"></a>00142 <span class="comment">//assign cache</span>
116<a name="l00143"></a>00143 <span class="comment">//est</span>
117<a name="l00144"></a>00144 _mu = <a class="code" href="classKalman.html#5568c74bac67ae6d3b1061dba60c9424" title="posterior density on $x_t$">est</a>._mu();
118<a name="l00145"></a>00145 <a class="code" href="classKalman.html#5568c74bac67ae6d3b1061dba60c9424" title="posterior density on $x_t$">est</a>._R(_P,_iP);
119<a name="l00146"></a>00146
120<a name="l00147"></a>00147 <span class="comment">//fy</span>
121<a name="l00148"></a>00148 _yp = <a class="code" href="classKalman.html#e580ab06483952bd03f2e651763e184f" title="preditive density on $y_t$">fy</a>._mu();
122<a name="l00149"></a>00149 <a class="code" href="classKalman.html#e580ab06483952bd03f2e651763e184f" title="preditive density on $y_t$">fy</a>._R(_Ry,_iRy);
123<a name="l00150"></a>00150 };
124<a name="l00151"></a>00151
125<a name="l00152"></a>00152 <span class="keyword">template</span>&lt;<span class="keyword">class</span> sq_T&gt;
126<a name="l00153"></a><a class="code" href="classKalman.html#239b28a0380946f5749b2f8d2807f93a">00153</a> <span class="keywordtype">void</span> <a class="code" href="classKalman.html#239b28a0380946f5749b2f8d2807f93a" title="Set parameters with check of relevance.">Kalman&lt;sq_T&gt;::set_parameters</a>(<span class="keyword">const</span> mat &amp;A0,<span class="keyword">const</span>  mat &amp;B0, <span class="keyword">const</span> mat &amp;C0, <span class="keyword">const</span> mat &amp;D0, <span class="keyword">const</span> sq_T &amp;R0, <span class="keyword">const</span> sq_T &amp;Q0) {
127<a name="l00154"></a>00154         it_assert_debug( A0.cols()==dimx, <span class="stringliteral">"Kalman: A is not square"</span> );
128<a name="l00155"></a>00155         it_assert_debug( B0.rows()==dimx, <span class="stringliteral">"Kalman: B is not compatible"</span> );
129<a name="l00156"></a>00156         it_assert_debug( C0.cols()==dimx, <span class="stringliteral">"Kalman: C is not square"</span> );
130<a name="l00157"></a>00157         it_assert_debug(( D0.rows()==dimy ) || ( D0.cols()==dimu ),     <span class="stringliteral">"Kalman: D is not compatible"</span> );
131<a name="l00158"></a>00158         it_assert_debug(( R0.cols()==dimy ) || ( R0.rows()==dimy ), <span class="stringliteral">"Kalman: R is not compatible"</span> );
132<a name="l00159"></a>00159         it_assert_debug(( Q0.cols()==dimx ) || ( Q0.rows()==dimx ), <span class="stringliteral">"Kalman: Q is not compatible"</span> );
133<a name="l00160"></a>00160
134<a name="l00161"></a>00161         A = A0;
135<a name="l00162"></a>00162         B = B0;
136<a name="l00163"></a>00163         C = C0;
137<a name="l00164"></a>00164         D = D0;
138<a name="l00165"></a>00165         R = R0;
139<a name="l00166"></a>00166         Q = Q0;
140<a name="l00167"></a>00167 }
141<a name="l00168"></a>00168
142<a name="l00169"></a>00169 <span class="keyword">template</span>&lt;<span class="keyword">class</span> sq_T&gt;
143<a name="l00170"></a><a class="code" href="classKalman.html#7750ffd73f261828a32c18aaeb65c75c">00170</a> <span class="keywordtype">void</span> <a class="code" href="classKalman.html#7750ffd73f261828a32c18aaeb65c75c" title="Here dt = [yt;ut] of appropriate dimensions.">Kalman&lt;sq_T&gt;::bayes</a>( <span class="keyword">const</span> vec &amp;dt) {
144<a name="l00171"></a>00171         it_assert_debug( dt.length()==( dimy+dimu ),<span class="stringliteral">"KalmanFull::bayes wrong size of dt"</span> );
145<a name="l00172"></a>00172
146<a name="l00173"></a>00173         vec u = dt.get( dimy,dimy+dimu-1 );
147<a name="l00174"></a>00174         vec y = dt.get( 0,dimy-1 );
148<a name="l00175"></a>00175         <span class="comment">//Time update</span>
149<a name="l00176"></a>00176         *_mu = A*(*_mu) + B*u;
150<a name="l00177"></a>00177         <span class="comment">//P  = A*P*A.transpose() + Q; in sq_T</span>
151<a name="l00178"></a>00178         _P-&gt;mult_sym( A );
152<a name="l00179"></a>00179         (*_P)+=Q;
153<a name="l00180"></a>00180
154<a name="l00181"></a>00181         <span class="comment">//Data update</span>
155<a name="l00182"></a>00182         <span class="comment">//_Ry = C*P*C.transpose() + R; in sq_T</span>
156<a name="l00183"></a>00183         _P-&gt;mult_sym( C, *_Ry);
157<a name="l00184"></a>00184         (*_Ry)+=R;
158<a name="l00185"></a>00185
159<a name="l00186"></a>00186         mat Pfull = _P-&gt;to_mat();
160<a name="l00187"></a>00187         
161<a name="l00188"></a>00188         _Ry-&gt;inv( *_iRy ); <span class="comment">// result is in _iRy;</span>
162<a name="l00189"></a>00189         _K = Pfull*C.transpose()*(_iRy-&gt;to_mat());
163<a name="l00190"></a>00190         
164<a name="l00191"></a>00191         sq_T pom((<span class="keywordtype">int</span>)Pfull.rows());
165<a name="l00192"></a>00192         _iRy-&gt;mult_sym_t(C*Pfull,pom);
166<a name="l00193"></a>00193         (*_P) -= pom; <span class="comment">// P = P -PC'iRy*CP;</span>
167<a name="l00194"></a>00194         (*_yp) = C*(*_mu)+D*u; <span class="comment">//y prediction</span>
168<a name="l00195"></a>00195         (*_mu) += _K*( y-(*_yp) );
169<a name="l00196"></a>00196         
170<a name="l00197"></a>00197         <span class="keywordflow">if</span> (<a class="code" href="classBM.html#bf6fb59b30141074f8ee1e2f43d03129" title="If true, the filter will compute likelihood of the data record and store it in ll...">evalll</a>==<span class="keyword">true</span>) {
171<a name="l00198"></a>00198                 <a class="code" href="classBM.html#5623fef6572a08c2b53b8c87b82dc979" title="Logarithm of marginalized data likelihood.">ll</a>+=<a class="code" href="classKalman.html#e580ab06483952bd03f2e651763e184f" title="preditive density on $y_t$">fy</a>.evalpdflog(*_yp);
172<a name="l00199"></a>00199         }
173<a name="l00200"></a>00200 };
174<a name="l00201"></a>00201
175<a name="l00202"></a>00202 <span class="comment">//TODO why not const pointer??</span>
176<a name="l00203"></a>00203
177<a name="l00204"></a>00204 <span class="keyword">template</span>&lt;<span class="keyword">class</span> sq_T&gt;
178<a name="l00205"></a><a class="code" href="classEKF.html#ea4f3254cacf0a92d2a820b1201d049e">00205</a> <a class="code" href="classEKF.html#ea4f3254cacf0a92d2a820b1201d049e" title="Default constructor.">EKF&lt;sq_T&gt;::EKF</a>(<a class="code" href="classRV.html" title="Class representing variables, most often random variables.">RV</a> rvx0, <a class="code" href="classRV.html" title="Class representing variables, most often random variables.">RV</a> rvy0, <a class="code" href="classRV.html" title="Class representing variables, most often random variables.">RV</a> rvu0): <a class="code" href="classKalman.html" title="Kalman filter with covariance matrices in square root form.">Kalman</a>&lt;ldmat&gt;(rvx0,rvy0,rvu0){}
179<a name="l00206"></a>00206
180<a name="l00207"></a>00207 <span class="keyword">template</span>&lt;<span class="keyword">class</span> sq_T&gt;
181<a name="l00208"></a>00208 <span class="keywordtype">void</span> <a class="code" href="classEKF.html" title="Extended Kalman Filter.">EKF&lt;sq_T&gt;::set_parameters</a>(<a class="code" href="classdiffbifn.html" title="Class representing a differentiable function of two variables $f(x,u)$.">diffbifn</a>* pfxu0,  <a class="code" href="classdiffbifn.html" title="Class representing a differentiable function of two variables $f(x,u)$.">diffbifn</a>* phxu0,<span class="keyword">const</span> sq_T Q0,<span class="keyword">const</span> sq_T R0) {
182<a name="l00209"></a>00209                 pfxu = pfxu0;
183<a name="l00210"></a>00210                 phxu = phxu0;
184<a name="l00211"></a>00211                 
185<a name="l00212"></a>00212                 <span class="comment">//initialize matrices A C, later, these will be only updated!</span>
186<a name="l00213"></a>00213                 pfxu-&gt;<a class="code" href="classdiffbifn.html#6d217a02d4fa13931258d4bebdd0feb4" title="Evaluates  and writes result into A .">dfdx_cond</a>(*_mu,zeros(dimu),A,<span class="keyword">true</span>);
187<a name="l00214"></a>00214                 pfxu-&gt;<a class="code" href="classdiffbifn.html#1978bafd7909d15c139a08c495c24aa0" title="Evaluates  and writes result into A .">dfdu_cond</a>(*_mu,zeros(dimu),B,<span class="keyword">true</span>);
188<a name="l00215"></a>00215                 phxu-&gt;<a class="code" href="classdiffbifn.html#6d217a02d4fa13931258d4bebdd0feb4" title="Evaluates  and writes result into A .">dfdx_cond</a>(*_mu,zeros(dimu),C,<span class="keyword">true</span>);
189<a name="l00216"></a>00216                 phxu-&gt;<a class="code" href="classdiffbifn.html#1978bafd7909d15c139a08c495c24aa0" title="Evaluates  and writes result into A .">dfdu_cond</a>(*_mu,zeros(dimu),D,<span class="keyword">true</span>);
190<a name="l00217"></a>00217
191<a name="l00218"></a>00218                 R = R0;
192<a name="l00219"></a>00219                 Q = Q0;
193<a name="l00220"></a>00220
194<a name="l00221"></a>00221         <span class="keyword">using</span> std::cout;
195<a name="l00222"></a>00222         cout&lt;&lt;A&lt;&lt;std::endl;
196<a name="l00223"></a>00223         cout&lt;&lt;B&lt;&lt;std::endl;
197<a name="l00224"></a>00224         cout&lt;&lt;C&lt;&lt;std::endl;
198<a name="l00225"></a>00225         cout&lt;&lt;D&lt;&lt;std::endl;
199<a name="l00226"></a>00226
200<a name="l00227"></a>00227 }
201<a name="l00228"></a>00228
202<a name="l00229"></a>00229 <span class="keyword">template</span>&lt;<span class="keyword">class</span> sq_T&gt;
203<a name="l00230"></a><a class="code" href="classEKF.html#c79c62c9b3e0b56b3aaa1b6f1d9a7af7">00230</a> <span class="keywordtype">void</span> <a class="code" href="classEKF.html#c79c62c9b3e0b56b3aaa1b6f1d9a7af7" title="Here dt = [yt;ut] of appropriate dimensions.">EKF&lt;sq_T&gt;::bayes</a>( <span class="keyword">const</span> vec &amp;dt) {
204<a name="l00231"></a>00231         it_assert_debug( dt.length()==( dimy+dimu ),<span class="stringliteral">"KalmanFull::bayes wrong size of dt"</span> );
205<a name="l00232"></a>00232
206<a name="l00233"></a>00233         vec u = dt.get( dimy,dimy+dimu-1 );
207<a name="l00234"></a>00234         vec y = dt.get( 0,dimy-1 );
208<a name="l00235"></a>00235         <span class="comment">//Time update</span>
209<a name="l00236"></a>00236         *_mu = pfxu-&gt;<a class="code" href="classdiffbifn.html#ad7673e16aa1a046b131b24c731c4632" title="Evaluates $f(x0,u0)$ (VS: Do we really need common eval? ).">eval</a>(*_mu, u);
210<a name="l00237"></a>00237         pfxu-&gt;<a class="code" href="classdiffbifn.html#6d217a02d4fa13931258d4bebdd0feb4" title="Evaluates  and writes result into A .">dfdx_cond</a>(*_mu,u,A,<span class="keyword">false</span>); <span class="comment">//update A by a derivative of fx</span>
211<a name="l00238"></a>00238         
212<a name="l00239"></a>00239         <span class="comment">//P  = A*P*A.transpose() + Q; in sq_T</span>
213<a name="l00240"></a>00240         _P-&gt;mult_sym( A );
214<a name="l00241"></a>00241         (*_P)+=Q;
215<a name="l00242"></a>00242
216<a name="l00243"></a>00243         <span class="comment">//Data update</span>
217<a name="l00244"></a>00244         phxu-&gt;<a class="code" href="classdiffbifn.html#6d217a02d4fa13931258d4bebdd0feb4" title="Evaluates  and writes result into A .">dfdx_cond</a>(*_mu,u,C,<span class="keyword">false</span>); <span class="comment">//update C by a derivative hx</span>
218<a name="l00245"></a>00245         <span class="comment">//_Ry = C*P*C.transpose() + R; in sq_T</span>
219<a name="l00246"></a>00246         _P-&gt;mult_sym( C, *_Ry);
220<a name="l00247"></a>00247         (*_Ry)+=R;
221<a name="l00248"></a>00248
222<a name="l00249"></a>00249         mat Pfull = _P-&gt;to_mat();
223<a name="l00250"></a>00250         
224<a name="l00251"></a>00251         _Ry-&gt;inv( *_iRy ); <span class="comment">// result is in _iRy;</span>
225<a name="l00252"></a>00252         _K = Pfull*C.transpose()*(_iRy-&gt;to_mat());
226<a name="l00253"></a>00253         
227<a name="l00254"></a>00254         sq_T pom((<span class="keywordtype">int</span>)Pfull.rows());
228<a name="l00255"></a>00255         _iRy-&gt;mult_sym_t(C*Pfull,pom);
229<a name="l00256"></a>00256         (*_P) -= pom; <span class="comment">// P = P -PC'iRy*CP;</span>
230<a name="l00257"></a>00257         *_yp = phxu-&gt;<a class="code" href="classdiffbifn.html#ad7673e16aa1a046b131b24c731c4632" title="Evaluates $f(x0,u0)$ (VS: Do we really need common eval? ).">eval</a>(*_mu,u); <span class="comment">//y prediction</span>
231<a name="l00258"></a>00258         (*_mu) += _K*( y-*_yp );
232<a name="l00259"></a>00259         
233<a name="l00260"></a>00260         <span class="keywordflow">if</span> (<a class="code" href="classBM.html#bf6fb59b30141074f8ee1e2f43d03129" title="If true, the filter will compute likelihood of the data record and store it in ll...">evalll</a>==<span class="keyword">true</span>) {<a class="code" href="classBM.html#5623fef6572a08c2b53b8c87b82dc979" title="Logarithm of marginalized data likelihood.">ll</a>+=<a class="code" href="classKalman.html#5568c74bac67ae6d3b1061dba60c9424" title="posterior density on $x_t$">est</a>.<a class="code" href="classenorm.html#9517594915e897584eaebbb057ed8881" title="Compute log-probability of argument val.">evalpdflog</a>(y);}
234<a name="l00261"></a>00261 };
235<a name="l00262"></a>00262
236<a name="l00263"></a>00263
237<a name="l00264"></a>00264 <span class="preprocessor">#endif // KF_H</span>
238<a name="l00265"></a>00265 <span class="preprocessor"></span>
239<a name="l00266"></a>00266
240</pre></div><hr size="1"><address style="text-align: right;"><small>Generated on Thu Feb 28 16:54:39 2008 for mixpp by&nbsp;
[8]241<a href="http://www.doxygen.org/index.html">
242<img src="doxygen.png" alt="doxygen" align="middle" border="0"></a> 1.5.3 </small></address>
243</body>
244</html>
Note: See TracBrowser for help on using the browser.