AdaptiveDMD.cpp

 
// Description: Implementation of the AdaptiveDMD algorithm.
 
#include "AdaptiveDMD.h"
#include "manifold_interp/VectorInterpolator.h"
 
#include "linalg/Matrix.h"
#include "linalg/Vector.h"
#include <algorithm>
 
namespace CAROM {
 
AdaptiveDMD::AdaptiveDMD(int dim, double desired_dt, std::string rbf,
                         std::string interp_method,
                         double closest_rbf_val,
                         bool alt_output_basis,
                         Vector* state_offset) :
    DMD(dim, alt_output_basis, state_offset)
{
    CAROM_VERIFY(rbf == "G" || rbf == "IQ" || rbf == "IMQ");
    CAROM_VERIFY(interp_method == "LS" || interp_method == "IDW"
                 || interp_method == "LP");
    CAROM_VERIFY(closest_rbf_val >= 0.0 && closest_rbf_val <= 1.0);
    d_dt = desired_dt;
    d_interp_method = interp_method;
    d_rbf = rbf;
    d_closest_rbf_val = closest_rbf_val;
}
 
AdaptiveDMD::~AdaptiveDMD()
{
    for (auto interp_snapshot : d_interp_snapshots)
    {
        delete interp_snapshot;
    }
}
 
void AdaptiveDMD::train(double energy_fraction, const Matrix* W0,
                        double linearity_tol)
{
    const Matrix* f_snapshots = getInterpolatedSnapshots();
    CAROM_VERIFY(f_snapshots->numColumns() > 1);
    CAROM_VERIFY(energy_fraction > 0 && energy_fraction <= 1);
    d_energy_fraction = energy_fraction;
    constructDMD(f_snapshots, d_rank, d_num_procs, W0, linearity_tol);
 
    delete f_snapshots;
}
 
void AdaptiveDMD::train(int k, const Matrix* W0, double linearity_tol)
{
    const Matrix* f_snapshots = getInterpolatedSnapshots();
    CAROM_VERIFY(f_snapshots->numColumns() > 1);
    CAROM_VERIFY(k > 0 && k <= f_snapshots->numColumns() - 1);
    d_energy_fraction = -1.0;
    d_k = k;
    constructDMD(f_snapshots, d_rank, d_num_procs, W0, linearity_tol);
 
    delete f_snapshots;
}
 
void AdaptiveDMD::interpolateSnapshots()
{
    CAROM_VERIFY(d_interp_snapshots.size() == 0);
    CAROM_VERIFY(d_snapshots.size() == d_sampled_times.size());
    CAROM_VERIFY(d_sampled_times.size() > 1);
 
    if (d_rank == 0) std::cout << "Number of snapshots is: " << d_snapshots.size()
                                   << std::endl;
 
    bool automate_dt = false;
    if (d_dt <= 0.0)
    {
        automate_dt = true;
        std::vector<double> d_sampled_dts;
        for (int i = 1; i < d_sampled_times.size(); i++)
        {
            d_sampled_dts.push_back(d_sampled_times[i]->item(0) - d_sampled_times[i -
                                    1]->item(0));
        }
 
        auto m = d_sampled_dts.begin() + d_sampled_dts.size() / 2;
        std::nth_element(d_sampled_dts.begin(), m, d_sampled_dts.end());
        if (d_rank == 0) std::cout <<
                                       "Setting desired dt to the median dt between the samples: " <<
                                       d_sampled_dts[d_sampled_dts.size() / 2] << std::endl;
        d_dt = d_sampled_dts[d_sampled_dts.size() / 2];
    }
    CAROM_VERIFY(d_sampled_times.back()->item(0) > d_dt);
 
    // Find the nearest dt that evenly divides the snapshots.
    int num_time_steps = std::round(d_sampled_times.back()->item(0) / d_dt);
    if (automate_dt && num_time_steps < d_sampled_times.size())
    {
        num_time_steps = d_sampled_times.size();
        if (d_rank == 0) std::cout <<
                                       "There will be less interpolated snapshots than FOM snapshots. dt will be decreased."
                                       << std::endl;
    }
    double new_dt = d_sampled_times.back()->item(0) / num_time_steps;
    if (new_dt != d_dt)
    {
        d_dt = new_dt;
        if (d_rank == 0) std::cout << "Setting desired dt to " << d_dt <<
                                       " to ensure a constant dt given the final sampled time." << std::endl;
    }
 
    // Solve the linear system if required.
    Matrix* f_T = NULL;
    double epsilon = convertClosestRBFToEpsilon(d_sampled_times, d_rbf,
                     d_closest_rbf_val);
    if (d_interp_method == "LS")
    {
        f_T = solveLinearSystem(d_sampled_times, d_snapshots, d_interp_method, d_rbf,
                                epsilon);
    }
 
    // Create interpolated snapshots using d_dt as the desired dt.
    for (int i = 0; i <= num_time_steps; i++)
    {
        double curr_time = i * d_dt;
        if (d_rank == 0) std::cout << "Creating new interpolated sample at: " <<
                                       d_t_offset + curr_time << std::endl;
        CAROM::Vector* point = new Vector(&curr_time, 1, false);
 
        // Obtain distances from database points to new point
        std::vector<double> rbf = obtainRBFToTrainingPoints(d_sampled_times,
                                  d_interp_method, d_rbf, epsilon, point);
 
        // Obtain the interpolated snapshot.
        CAROM::Vector* curr_interpolated_snapshot = obtainInterpolatedVector(
                    d_snapshots, f_T, d_interp_method, rbf);
        d_interp_snapshots.push_back(curr_interpolated_snapshot);
 
        delete point;
    }
 
    if (d_rank == 0) std::cout << "Number of interpolated snapshots is: " <<
                                   d_interp_snapshots.size() << std::endl;
}
 
double AdaptiveDMD::getTrueDt() const
{
    return d_dt;
}
 
const Matrix* AdaptiveDMD::getInterpolatedSnapshots()
{
    if (d_interp_snapshots.size() == 0) interpolateSnapshots();
    return createSnapshotMatrix(d_interp_snapshots);
}
 
}