data_driven_instationary

pymor-demo data_driven_instationary [OPTIONS] PROBLEM_NUMBER REGRESSOR GRID_INTERVALS TIME_STEPS TRAINING_SAMPLES

Model order reduction with machine learning methods for instationary problems.

Problem number 0 considers the incompressible Navier-Stokes equations in a two-dimensional cavity with the Reynolds number as parameter. The discretization is based on FEniCS.

Problem number 1 considers a parametrized Burgers equation on a one-dimensional domain. The discretization is based on pyMOR’s built-in functionality.

Arguments:

PROBLEM_NUMBER

Selects the problem to solve [0 or 1]. [Required, Choices: 0, 1]

REGRESSOR

Regressor to use. Options are neural networks using PyTorch, pyMOR’s VKOGA algorithm or Gaussian process regression using scikit-learn. [Required, Choices: fcnn, vkoga, gpr]

GRID_INTERVALS

Grid interval count. [Required]

TIME_STEPS

Number of time steps used for discretization. [Required]

TRAINING_SAMPLES

Number of samples used for computing the reduced basis and training the regressor. [Required]

Parameters:

--fv, --no-fv

Use finite volume discretization instead of finite elements. [Default: False]

--vis, --no-vis

Visualize full order solution and reduced solution for a test set. [Default: False]

--validation-ratio

Ratio of training data used for validation of the neural networks. [Default: 0.1]

--time-vectorized, --no-time-vectorized

Predict the whole time trajectory at once or iteratively. [Default: False]

--input-scaling, --no-input-scaling

Scale the input of the regressor (i.e. the parameter). [Default: False]

--output-scaling, --no-output-scaling

Scale the output of the regressor (i.e. reduced coefficients or output quantity). [Default: False]