lasdi.gplasdi

Classes

BayesianGLaSDI

This class runs a full GPLaSDI training. It takes into input the autoencoder defined as a PyTorch object and the

Functions

average_rom(autoencoder, physics, latent_dynamics, ...)
sample_roms(autoencoder, physics, latent_dynamics, ...)	Collect n_samples of ROM trajectories on param_grid.
get_fom_max_std(autoencoder, Zis)	Computes the maximum standard deviation accross the parameter space grid and finds the corresponding parameter location
optimizer_to(optim, device)

Module Contents

lasdi.gplasdi.average_rom(autoencoder, physics, latent_dynamics, gp_dictionary, param_grid)

lasdi.gplasdi.sample_roms(autoencoder, physics, latent_dynamics, gp_dictionary, param_grid, n_samples)

Collect n_samples of ROM trajectories on param_grid. gp_dictionary: list of Gaussian process regressors (size of n_test) param_grid: numpy 2d array n_samples: integer assert(len(gp_dictionnary) == param_grid.shape[0])

output: np.array of size [n_test, n_samples, physics.nt, autoencoder.n_z]

lasdi.gplasdi.get_fom_max_std(autoencoder, Zis)

Computes the maximum standard deviation accross the parameter space grid and finds the corresponding parameter location

lasdi.gplasdi.optimizer_to(optim, device)

class lasdi.gplasdi.BayesianGLaSDI(physics, autoencoder, latent_dynamics, param_space, config)

This class runs a full GPLaSDI training. It takes into input the autoencoder defined as a PyTorch object and the dictionnary containing all the training parameters. The “train” method with run the active learning training loop, compute the reconstruction and SINDy loss, train the GPs, and sample a new FOM data point.

X_train

X_test

autoencoder

latent_dynamics

physics

param_space

timer

n_samples

lr

n_iter

max_iter

max_greedy_iter

ld_weight

coef_weight

optimizer

MSE

path_checkpoint

path_results

best_loss

best_coefs = None

restart_iter = 0

training_loss = []

ae_loss = []

ld_loss = []

coef_loss = []

train()

get_new_sample_point()

export()

load(dict_)