Skip to content

Input Guide

We explain in detail the parameters listed in the JSON files in example/simple examples.

General Parameters

  • equ_type (string): Indicates the type of equation. This type must be registered in the ODEFACTORY, see Customize ODE for more details.
  • ode_solver:
    • type (string): Indicates the type of ODE solver. This type must be registered in the ODESOLVERFACTORY, see Customize ODE Solver for more details.
    • dt (float): Time step \(k\) of the ODE Solver.
    • t_step (float): Time step of the trajectories.
  • dataset:
    • n_traj (int): Number of trajectories.
    • traj_len (int): Length of each trajectory.
    • x_min/x_max (float or list): Range of the states. If a float is given, it means every state has the same range.
    • seed_x (int): Seed for generating the states.
  • dictionary:
    • dim_output (int): Output dimension \(N_{\Psi}\) of the dictionary.

EDMD-RBF Algorithm

  • dataset:
    • param (list): Fixed parameters \(u\) of the dataset.
  • dictionary:
    • reg (float): Regularization parameter \(\lambda\) for generating the RBF functions.

EDMDDL Algorithm

  • dataset:
    • param (list): Fixed parameters \(u\) of the dataset.
    • train_ratio (float): Ratio of the training set.
  • dictionary
    • dic_layer_sizes (list): Hidden layer sizes of the dictionary.
  • solver:
    • reg (float): Regularization parameter \(\lambda\) when training the Koopman operator.
    • reg_final (float): Regularization parameter \(\lambda\) when outputting the Koopman operator.
    • n_epochs (int): Number of training epochs.
    • batch_size (int): Batch size for training.
    • tol (float): Tolerance of training.
    • dic_lr (float): Learning rate for training the dictionary.

Parametric Koopman Learning

  • dataset
    • n_traj_per_param (int): Number of trajectories per parameter setting.
    • param_min/param_max (list): Range of the parameters.
    • seed_param (int): Seed for generating the parameters.
    • param_time_dependent (bool): Indicates whether the parameters are time-dependent.
  • dictionary:
    • dic_layer_sizes (list): Hidden layer sizes of the dictionary.
  • solver:
    • n_epochs (int): Number of training epochs.
    • batch_size (int): Batch size for training.
    • tol (float): Tolerance of training.
    • dic_lr (float): Learning rate for training the dictionary.
    • koopman_layer_sizes (list): Hidden layer sizes for the Koopman operator.
    • koopman_lr (float): Learning rate for training the Koopman operator.