neural_operators.data#

class neural_operators.data.dataMethods.DataHandler(X_train_, X_trunk_, Y_train_, convert_to_tensor=True)[source]#

Bases: Dataset

torch Dataset for the (X_train, X_trunk, Y_train) triple. X_trunk is shared across every sample, or omitted from __getitem__ entirely when it’s None (PCANet has no trunk net).

convert_np_to_tensor(array)[source]#
class neural_operators.data.dataMethods.DataProcessor(data_file_name='../problems/poisson/data/Poisson_samples.npz', num_train=1900, num_test=100, num_inp_fn_points=2601, num_out_fn_points=2601, num_Y_components=1, num_inp_red_dim=None, num_out_red_dim=None)[source]#

Bases: object

Loads a .npz sample file, splits it into train/test, standardizes, and (for PCANet) optionally SVD-reduces each side to num_inp_red_dim/ num_out_red_dim. Use encoder_X/decoder_X and the _Y counterparts to go between physical and model space, not the raw SVD helpers.

compute_svd(data, num_red_dim, is_data_centered=False)[source]#
decoder_X(x)[source]#
decoder_Y(x)[source]#
encoder_X(x)[source]#
encoder_Y(x)[source]#
get_data_to_save()[source]#
lift_SVD(data, Pi)[source]#
load_X_data(data)[source]#
load_Y_data(data)[source]#
project_SVD(data, Pi)[source]#
class neural_operators.data.dataMethods.DataProcessorFNO(data_file_name='../problems/poisson/data/Poisson_FNO_samples.npz', num_train=1900, num_test=100, num_Y_components=1, coarsen_grid_factor=2)[source]#

Bases: object

DataProcessor’s counterpart for FNO: loads grid sample arrays, coarsens by coarsen_grid_factor, folds grid coordinates into X_train/X_test as extra channels. No SVD reduction.

decoder_X(x)[source]#
decoder_Y(x)[source]#
encoder_X(x)[source]#
encoder_Y(x)[source]#
get_data_to_save()[source]#
load_X_data(data, tol=1e-09)[source]#
load_Y_data(data, tol=1e-09)[source]#
class neural_operators.data.dataMethods.DataProcessorTF(batch_size=100, data_file_name='../problems/poisson/data/Poisson_samples.npz', num_train=1900, num_test=100, num_inp_fn_points=2601, num_out_fn_points=2601, num_Y_components=1, num_inp_red_dim=None, num_out_red_dim=None)[source]#

Bases: DataProcessor

DataProcessor reshaped for TensorFlow-style DeepONet layers (adds a singleton axis, overrides the encoders accordingly).

decoder_X(x)[source]#
decoder_Y(x)[source]#
encoder_X(x)[source]#
encoder_Y(x)[source]#
get_data_to_save()[source]#
minibatch()[source]#
testbatch(num_test)[source]#