Parametric_tw_csv example

[dreamer2368]

Miscellaneous addition to pylibROM

• Bindings for utils::CSVDatabase and utils::HDFDatabase
• python_utils/cpp_utils.hpp now provides a template function get1DArrayFromPtr for numpy array.
  • This resolves the issue Bindings with STL class input/output #11 , where the view numpy array is returned instead of the actual std::vector output variable.
  • The example is in line 110 of utils/pyCSVDatabase.cpp:

    csvdb.def("getIntegerVector", [](
        CSVDatabase &self, const std::string& key, bool append)
    {
        std::vector<int> *datavec = new std::vector<int>;
        self.getIntegerVector(key, *datavec, append);
        return get1DArrayFromPtr(datavec->data(), datavec->size(), true);
    },

examples/dmd/parametric_tw_csv.py example demonstration

parametric_tw_csv.py demonstrates the parametric DMD, along with the use of CSV/HDF databases.

1. FOM sampling process

• csv format

bash ./heat_conduction_csv.sh

• hdf5 format

bash ./heat_conduction_hdf.sh

2. Case1: serial DMD
   Add -hdf tag if you're following the workflow for hdf5.

python3 parametric_tw_csv.py -o hc_parametric_serial -rdim 16 -dtc 0.01 -offline
python3 parametric_tw_csv.py -o hc_parametric_serial -rdim 16 -dtc 0.01 -online

3. Case2: time-windowing DMD
   Add -hdf tag if you're following the workflow for hdf5.

python3 parametric_tw_csv.py -o hc_parametric_tw -nwinsamp 25 -dtc 0.01 -offline
python3 parametric_tw_csv.py -o hc_parametric_tw -nwinsamp 25 -dtc 0.01 -offline

Demonstration results

1. serial DMD, csv

Norm of true solution at t = 0.500

000 is 84.539879
Absolute error of DMD solution at t = 0.500000 is 0.106531
Relative error of DMD solution at t = 0.500000 is 0.001260

Elapsed time for training DMD: 0.000000e+00 second
Elapsed time for preprocessing DMD: 2.193182e-01 second
Total elapsed time for predicting DMD: 1.743942e+00 second
Average elapsed time for predicting DMD: 3.419494e-02 second

little difference from c++ libROM:

Norm of true solution at t = 0.5000000000000002 is 84.53987904150704
Absolute error of DMD solution at t = 0.5000000000000002 is 0.1065308142113612
Relative error of DMD solution at t = 0.5000000000000002 is 0.001260124989758469
Elapsed time for training DMD: 0.000000e+00 second
Elapsed time for preprocessing DMD: 2.209829e-01 second
Total elapsed time for predicting DMD: 1.791613e+00 second
Average elapsed time for predicting DMD: 3.512967e-02 second

2. serial DMD, hdf

Norm of true solution at t = 0.500000 is 84.539879
Absolute error of DMD solution at t = 0.500000 is 0.106531
Relative error of DMD solution at t = 0.500000 is 0.001260

Elapsed time for training DMD: 0.000000e+00 second
Elapsed time for preprocessing DMD: 2.122204e-01 second
Total elapsed time for predicting DMD: 1.759935e+00 second
Average elapsed time for predicting DMD: 3.450852e-02 second

similar result from c++ libROM:

Norm of true solution at t = 0.5000000000000002 is 84.53987904150704
Absolute error of DMD solution at t = 0.5000000000000002 is 0.1065308142113612
Relative error of DMD solution at t = 0.5000000000000002 is 0.001260124989758469
Elapsed time for training DMD: 0.000000e+00 second
Elapsed time for preprocessing DMD: 1.646451e-01 second
Total elapsed time for predicting DMD: 1.757581e+00 second
Average elapsed time for predicting DMD: 3.446237e-02 second

3. time-windowing, csv

Norm of true solution at t = 0.500000 is 84.539879
Absolute error of DMD solution at t = 0.500000 is 0.055013
Relative error of DMD solution at t = 0.500000 is 0.000651

Elapsed time for training DMD: 0.000000e+00 second
Elapsed time for preprocessing DMD: 1.072407e-01 second
Total elapsed time for predicting DMD: 2.882011e-01 second
Average elapsed time for predicting DMD: 5.651002e-03 second

c++ libROM:

Norm of true solution at t = 0.5000000000000002 is 84.53987904150704
Absolute error of DMD solution at t = 0.5000000000000002 is 0.0550131373888892
Relative error of DMD solution at t = 0.5000000000000002 is 0.0006507359368456049
Elapsed time for training DMD: 0.000000e+00 second
Elapsed time for preprocessing DMD: 8.679060e-02 second
Total elapsed time for predicting DMD: 2.865902e-01 second
Average elapsed time for predicting DMD: 5.619415e-03 second

4. time-windowing, hdf

Norm of true solution at t = 0.500000 is 84.539879
Absolute error of DMD solution at t = 0.500000 is 0.055013
Relative error of DMD solution at t = 0.500000 is 0.000651

Elapsed time for training DMD: 0.000000e+00 second
Elapsed time for preprocessing DMD: 1.108568e-01 second
Total elapsed time for predicting DMD: 3.017945e-01 second
Average elapsed time for predicting DMD: 5.917540e-03 second

c++ libROM:

Norm of true solution at t = 0.5000000000000002 is 84.53987904150704
Absolute error of DMD solution at t = 0.5000000000000002 is 0.05501313738888904
Relative error of DMD solution at t = 0.5000000000000002 is 0.000650735936845603
Elapsed time for training DMD: 0.000000e+00 second
Elapsed time for preprocessing DMD: 7.375104e-02 second
Total elapsed time for predicting DMD: 2.931510e-01 second
Average elapsed time for predicting DMD: 5.748059e-03 second