sisr_filter.h

Go to the documentation of this file.

#ifndef SISR_FILTER_H
#define SISR_FILTER_H
 
#include <array>
 
#ifdef DROPPINGTHISINRPACKAGE
    #include <RcppEigen.h>
    // [[Rcpp::depends(RcppEigen)]]
#else
    #include <Eigen/Dense>
#endif
 
#include "pf_base.h"
 
 
namespace pf {
 
namespace filters {
 
 
template<size_t nparts, size_t dimx, size_t dimy, typename resamp_t, typename float_t, bool debug=false>
class SISRFilter : public bases::pf_base<float_t, dimy, dimx>
{
private:
 
    using ssv         = Eigen::Matrix<float_t, dimx, 1>; 
    using osv         = Eigen::Matrix<float_t, dimy, 1>; // obs size vec
    using Mat         = Eigen::Matrix<float_t,Eigen::Dynamic,Eigen::Dynamic>;
    using arrayStates = std::array<ssv, nparts>;
    using arrayfloat_t = std::array<float_t, nparts>;
   
public:
 
    SISRFilter(const unsigned int &rs=1);
    
    
    virtual ~SISRFilter();
    
    
    float_t getLogCondLike() const; 
    
    
    std::vector<Mat> getExpectations() const;
    
    
    void filter(const osv &data, const std::vector<std::function<const Mat(const ssv&)> >& fs = std::vector<std::function<const Mat(const ssv&)> >());
    
    
    virtual float_t logMuEv (const ssv &x1) = 0;
    
    
    virtual ssv q1Samp (const osv &y1) = 0;    
    
    
    virtual float_t logQ1Ev (const ssv &x1, const osv &y1 ) = 0;
    
    
    virtual float_t logGEv (const osv &yt, const ssv &xt ) = 0;
    
    
    virtual float_t logFEv (const ssv &xt, const ssv &xtm1 ) = 0;
    
    
    virtual ssv qSamp (const ssv &xtm1, const osv &yt ) = 0;
    
    
    virtual float_t logQEv (const ssv &xt, const ssv &xtm1, const osv &yt ) = 0;    
    
protected:
 
    arrayStates m_particles;
 
    arrayfloat_t m_logUnNormWeights;
    
    unsigned int m_now; 
 
    float_t m_logLastCondLike;  
    
    resamp_t m_resampler;
    
    std::vector<Mat> m_expectations; // stores any sample averages the user wants
    
    unsigned int m_resampSched;
    
    
};
 
 
 
template<size_t nparts, size_t dimx, size_t dimy, typename resamp_t, typename float_t, bool debug>
SISRFilter<nparts,dimx,dimy,resamp_t,float_t, debug>::SISRFilter(const unsigned int &rs)
                : m_now(0)
                , m_logLastCondLike(0.0)
                , m_resampSched(rs) 
{
    std::fill(m_logUnNormWeights.begin(), m_logUnNormWeights.end(), 0.0); // log(1) = 0
}
 
 
template<size_t nparts, size_t dimx, size_t dimy, typename resamp_t, typename float_t, bool debug>
SISRFilter<nparts,dimx,dimy,resamp_t,float_t, debug>::~SISRFilter() {}
 
    
template<size_t nparts, size_t dimx, size_t dimy, typename resamp_t, typename float_t, bool debug>
float_t SISRFilter<nparts,dimx,dimy,resamp_t,float_t, debug>::getLogCondLike() const
{
    return m_logLastCondLike;
}
    
 
template<size_t nparts, size_t dimx, size_t dimy, typename resamp_t, typename float_t, bool debug>    
auto SISRFilter<nparts,dimx,dimy,resamp_t,float_t, debug>::getExpectations() const -> std::vector<Mat> 
{
    return m_expectations;
}
 
 
template<size_t nparts, size_t dimx, size_t dimy, typename resamp_t, typename float_t, bool debug>
void SISRFilter<nparts,dimx,dimy,resamp_t,float_t, debug>::filter(const osv &data, const std::vector<std::function<const Mat(const ssv&)> >& fs)
{
 
    if(m_now > 0)
    {
 
        // try to iterate over particles all at once
        ssv newSamp;
        arrayfloat_t oldLogUnNormWts = m_logUnNormWeights;
        float_t maxOldLogUnNormWts(-std::numeric_limits<float_t>::infinity());
        for(size_t ii = 0; ii < nparts; ++ii)
        {
 
            // update max of old logUnNormWts before you change the element
            if (m_logUnNormWeights[ii] > maxOldLogUnNormWts)
                maxOldLogUnNormWts = m_logUnNormWeights[ii];
            
            // sample and get weight adjustments
            newSamp = qSamp(m_particles[ii], data);
            m_logUnNormWeights[ii] += logFEv(newSamp, m_particles[ii]);
            m_logUnNormWeights[ii] += logGEv(data, newSamp);
            m_logUnNormWeights[ii] -= logQEv(newSamp, m_particles[ii], data);
 
            // overwrite stuff
            m_particles[ii] = newSamp;
 
            #ifndef DROPPINGTHISINRPACKAGE
            if constexpr(debug) 
                std::cout << "time: " << m_now << ", transposed sample: " << m_particles[ii].transpose() << ", log unnorm weight: " << m_logUnNormWeights[ii] << "\n";
            #endif
 
        }
       
        // compute estimate of log p(y_t|y_{1:t-1}) with log-exp-sum trick
        float_t maxNumer = *std::max_element(m_logUnNormWeights.begin(), m_logUnNormWeights.end()); //because you added log adjustments
        float_t sumExp1(0.0);
        float_t sumExp2(0.0);
        for(size_t i = 0; i < nparts; ++i){
            sumExp1 += std::exp(m_logUnNormWeights[i] - maxNumer);
            sumExp2 += std::exp(oldLogUnNormWts[i] - maxOldLogUnNormWts);
        }
        m_logLastCondLike = maxNumer + std::log(sumExp1) - maxOldLogUnNormWts - std::log(sumExp2);
 
        // calculate expectations before you resample
        unsigned int fId(0);
        float_t weightNormConst(0.0);
        for(auto & h : fs){ // iterate over all functions
 
            Mat testOut = h(m_particles[0]);
            unsigned int rows = testOut.rows();
            unsigned int cols = testOut.cols();
            Mat numer = Mat::Zero(rows,cols);
            float_t denom(0.0);
 
            for(size_t prtcl = 0; prtcl < nparts; ++prtcl){ // iterate over all particles
                numer += h(m_particles[prtcl]) * std::exp(m_logUnNormWeights[prtcl] - maxNumer );
                denom += std::exp(m_logUnNormWeights[prtcl] - maxNumer);
            }
            m_expectations[fId] = numer/denom;
 
            // print stuff if debug mode is on
            #ifndef DROPPINGTHISINRPACKAGE
            if constexpr(debug)
                std::cout << "transposed expectation " << fId << ": " << m_expectations[fId].transpose() << "\n";
            #endif
 
            fId++;
        }
 
        // resample if you should
        if( (m_now + 1) % m_resampSched == 0)
            m_resampler.resampLogWts(m_particles, m_logUnNormWeights);
 
        // advance time
        m_now += 1;       
    }
    else // (m_now == 0) //time 1
    {
       
        // only need to iterate over particles once
        float_t sumWts(0.0);
        for(size_t ii = 0; ii < nparts; ++ii)
        {
            // sample particles
            m_particles[ii] = q1Samp(data);
            m_logUnNormWeights[ii] += logMuEv(m_particles[ii]);
            m_logUnNormWeights[ii] += logGEv(data, m_particles[ii]);
            m_logUnNormWeights[ii] -= logQ1Ev(m_particles[ii], data);
 
            #ifndef DROPPINGTHISINRPACKAGE
            if constexpr(debug) 
                std::cout << "time: " << m_now << ", transposed sample: " << m_particles[ii].transpose() << ", log unnorm weight: " << m_logUnNormWeights[ii] << "\n";
            #endif
 
 
        }
       
        // calculate log cond likelihood with log-exp-sum trick
        float_t max = *std::max_element(m_logUnNormWeights.begin(), m_logUnNormWeights.end());
        float_t sumExp(0.0);
        for(size_t i = 0; i < nparts; ++i){
            sumExp += std::exp(m_logUnNormWeights[i] - max);
        }
        m_logLastCondLike = -std::log(nparts) + max + std::log(sumExp);
   
        // calculate expectations before you resample
        m_expectations.resize(fs.size());
        unsigned int fId(0);
        for(auto & h : fs){
            
            Mat testOut = h(m_particles[0]);
            unsigned int rows = testOut.rows();
            unsigned int cols = testOut.cols();
            Mat numer = Mat::Zero(rows,cols);
            float_t denom(0.0);
 
            for(size_t prtcl = 0; prtcl < nparts; ++prtcl){ 
                numer += h(m_particles[prtcl]) * std::exp(m_logUnNormWeights[prtcl] - max);
                denom += std::exp(m_logUnNormWeights[prtcl] - max);
            }
            m_expectations[fId] = numer/denom;
 
            // print stuff if debug mode is on
            #ifndef DROPPINGTHISINRPACKAGE
            if constexpr(debug)
                std::cout << "transposed expectation " << fId << ": " << m_expectations[fId].transpose() << "\n";
            #endif
 
            fId++;
        }
   
        // resample if you should
        if( (m_now + 1) % m_resampSched == 0)
            m_resampler.resampLogWts(m_particles, m_logUnNormWeights);
   
        // advance time step
        m_now += 1;   
    }
 
}
 
 
 
template<size_t nparts, size_t dimx, size_t dimy, size_t dimu, size_t dimur, typename resamp_t, typename float_t, bool debug=false>
class SISRFilterCRN : public bases::pf_base_crn<float_t, dimy, dimx, dimu, dimur, nparts>
{
private:
 
    using ssv         = Eigen::Matrix<float_t, dimx, 1>; 
    using osv         = Eigen::Matrix<float_t, dimy, 1>; // obs size vec
    using usv         = Eigen::Matrix<float_t, dimu, 1>; 
    using usvr         = Eigen::Matrix<float_t, dimur, 1>; 
    using Mat         = Eigen::Matrix<float_t,Eigen::Dynamic,Eigen::Dynamic>;
    using arrayStates = std::array<ssv, nparts>;
    using arrayUs     = std::array<usv, nparts>;
    using arrayfloat_t = std::array<float_t, nparts>;
   
public:
 
    SISRFilterCRN(const unsigned int &rs=1);
    
    
    virtual ~SISRFilterCRN();
    
    
    float_t getLogCondLike() const; 
    
    
    std::vector<Mat> getExpectations() const;
    
    
    void filter(const osv &data, const arrayUs& Uarr, const usvr& Uresamp, const std::vector<std::function<const Mat(const ssv&)> >& fs = std::vector<std::function<const Mat(const ssv&)> >());
    
    
    virtual float_t logMuEv (const ssv &x1) = 0;
    
    
    virtual ssv Xi1 (const usv& U, const osv &y1) = 0;    
    
    
    virtual float_t logQ1Ev (const ssv &x1, const osv &y1 ) = 0;
    
    
    virtual float_t logGEv (const osv &yt, const ssv &xt ) = 0;
    
    
    virtual float_t logFEv (const ssv &xt, const ssv &xtm1 ) = 0;
    
    
    virtual ssv Xit (const ssv &xtm1, const usv& U, const osv &yt) = 0;
    
    
    virtual float_t logQEv (const ssv &xt, const ssv &xtm1, const osv &yt ) = 0;    
    
protected:
 
    arrayStates m_particles;
 
    arrayfloat_t m_logUnNormWeights;
    
    unsigned int m_now; 
 
    float_t m_logLastCondLike;  
    
    resamp_t m_resampler;
    
    std::vector<Mat> m_expectations; // stores any sample averages the user wants
    
    unsigned int m_resampSched;
    
 
};
 
 
template<size_t nparts, size_t dimx, size_t dimy, size_t dimu, size_t dimur, typename resamp_t, typename float_t, bool debug>
SISRFilterCRN<nparts,dimx,dimy,dimu,dimur,resamp_t,float_t, debug>::SISRFilterCRN(const unsigned int &rs)
                : m_now(0)
                , m_logLastCondLike(0.0)
                , m_resampSched(rs) 
{
    std::fill(m_logUnNormWeights.begin(), m_logUnNormWeights.end(), 0.0); // log(1) = 0
}
 
 
template<size_t nparts, size_t dimx, size_t dimy, size_t dimu, size_t dimur, typename resamp_t, typename float_t, bool debug>
SISRFilterCRN<nparts,dimx,dimy,dimu,dimur,resamp_t,float_t, debug>::~SISRFilterCRN() {}
 
 
template<size_t nparts, size_t dimx, size_t dimy, size_t dimu, size_t dimur, typename resamp_t, typename float_t, bool debug>
float_t SISRFilterCRN<nparts,dimx,dimy,dimu,dimur,resamp_t,float_t, debug>::getLogCondLike() const
{
    return m_logLastCondLike;
}
   
 
template<size_t nparts, size_t dimx, size_t dimy, size_t dimu, size_t dimur, typename resamp_t, typename float_t, bool debug>
auto SISRFilterCRN<nparts,dimx,dimy,dimu,dimur,resamp_t,float_t, debug>::getExpectations() const -> std::vector<Mat> 
{
    return m_expectations;
}
 
 
template<size_t nparts, size_t dimx, size_t dimy, size_t dimu, size_t dimur, typename resamp_t, typename float_t, bool debug>
void SISRFilterCRN<nparts,dimx,dimy,dimu,dimur,resamp_t,float_t, debug>::filter(const osv &data, const arrayUs& Uarr, const usvr& Uresamp, const std::vector<std::function<const Mat(const ssv&)> >& fs)
{
 
    if(m_now > 0)
    {
 
        // try to iterate over particles all at once
        ssv newSamp;
        arrayfloat_t oldLogUnNormWts = m_logUnNormWeights;
        float_t maxOldLogUnNormWts(-std::numeric_limits<float_t>::infinity());
        for(size_t ii = 0; ii < nparts; ++ii)
        {
 
            // update max of old logUnNormWts before you change the element
            if (m_logUnNormWeights[ii] > maxOldLogUnNormWts)
                maxOldLogUnNormWts = m_logUnNormWeights[ii];
            
            // sample and get weight adjustments
            newSamp = Xit(m_particles[ii], Uarr[ii], data);
            m_logUnNormWeights[ii] += logFEv(newSamp, m_particles[ii]);
            m_logUnNormWeights[ii] += logGEv(data, newSamp);
            m_logUnNormWeights[ii] -= logQEv(newSamp, m_particles[ii], data);
 
            // overwrite stuff
            m_particles[ii] = newSamp;
 
            #ifndef DROPPINGTHISINRPACKAGE
            if constexpr(debug) 
                std::cout << "time: " << m_now << ", transposed sample: " << m_particles[ii].transpose() << ", log unnorm weight: " << m_logUnNormWeights[ii] << "\n";
            #endif
        }
       
        // compute estimate of log p(y_t|y_{1:t-1}) with log-exp-sum trick
        float_t maxNumer = *std::max_element(m_logUnNormWeights.begin(), m_logUnNormWeights.end()); //because you added log adjustments
        float_t sumExp1(0.0);
        float_t sumExp2(0.0);
        for(size_t i = 0; i < nparts; ++i){
            sumExp1 += std::exp(m_logUnNormWeights[i] - maxNumer);
            sumExp2 += std::exp(oldLogUnNormWts[i] - maxOldLogUnNormWts);
        }
        m_logLastCondLike = maxNumer + std::log(sumExp1) - maxOldLogUnNormWts - std::log(sumExp2);
 
        // calculate expectations before you resample
        unsigned int fId(0);
        float_t weightNormConst(0.0);
        for(auto & h : fs){ // iterate over all functions
 
            Mat testOut = h(m_particles[0]);
            unsigned int rows = testOut.rows();
            unsigned int cols = testOut.cols();
            Mat numer = Mat::Zero(rows,cols);
            float_t denom(0.0);
 
            for(size_t prtcl = 0; prtcl < nparts; ++prtcl){ // iterate over all particles
                numer += h(m_particles[prtcl]) * std::exp(m_logUnNormWeights[prtcl] - maxNumer );
                denom += std::exp(m_logUnNormWeights[prtcl] - maxNumer);
            }
            m_expectations[fId] = numer/denom;
 
            // print stuff if debug mode is on
            #ifndef DROPPINGTHISINRPACKAGE
            if constexpr(debug)
                std::cout << "transposed expectation " << fId << ": " << m_expectations[fId].transpose() << "\n";
            #endif
 
            fId++;
        }
 
        // resample if you should
        if( (m_now + 1) % m_resampSched == 0)
            m_resampler.resampLogWts(m_particles, m_logUnNormWeights, Uresamp);
 
        // advance time
        m_now += 1;       
    }
    else // (m_now == 0) //time 1
    {
       
        // only need to iterate over particles once
        float_t sumWts(0.0);
        for(size_t ii = 0; ii < nparts; ++ii)
        {
            // sample particles
            m_particles[ii] = Xi1(Uarr[ii], data);
            m_logUnNormWeights[ii] += logMuEv(m_particles[ii]);
            m_logUnNormWeights[ii] += logGEv(data, m_particles[ii]);
            m_logUnNormWeights[ii] -= logQ1Ev(m_particles[ii], data);
 
            #ifndef DROPPINGTHISINRPACKAGE
            if constexpr(debug) 
                std::cout << "time: " << m_now << ", transposed sample: " << m_particles[ii].transpose() << ", log unnorm weight: " << m_logUnNormWeights[ii] << "\n";
            #endif
 
        }
       
        // calculate log cond likelihood with log-exp-sum trick
        float_t max = *std::max_element(m_logUnNormWeights.begin(), m_logUnNormWeights.end());
        float_t sumExp(0.0);
        for(size_t i = 0; i < nparts; ++i){
            sumExp += std::exp(m_logUnNormWeights[i] - max);
        }
        m_logLastCondLike = -std::log(nparts) + max + std::log(sumExp);
   
        // calculate expectations before you resample
        m_expectations.resize(fs.size());
        unsigned int fId(0);
        for(auto & h : fs){
            
            Mat testOut = h(m_particles[0]);
            unsigned int rows = testOut.rows();
            unsigned int cols = testOut.cols();
            Mat numer = Mat::Zero(rows,cols);
            float_t denom(0.0);
 
            for(size_t prtcl = 0; prtcl < nparts; ++prtcl){ 
                numer += h(m_particles[prtcl]) * std::exp(m_logUnNormWeights[prtcl] - max);
                denom += std::exp(m_logUnNormWeights[prtcl] - max);
            }
            m_expectations[fId] = numer/denom;
 
            // print stuff if debug mode is on
            #ifndef DROPPINGTHISINRPACKAGE
            if constexpr(debug)
                std::cout << "transposed expectation " << fId << ": " << m_expectations[fId].transpose() << "\n";
            #endif
 
            fId++;
        }
   
        // resample if you should
        if( (m_now + 1) % m_resampSched == 0)
            m_resampler.resampLogWts(m_particles, m_logUnNormWeights, Uresamp);
   
        // advance time step
        m_now += 1;   
    }
 
}
 
} // namespace filters
} //namespace pf
 
#endif //SISR_FILTER_H