def __init__(self, weights=None):
'''
Constructor for the class LinearModelLearningSquareLoss.
When approximating a vector-valued function, setting sharedCoefficients
to false will independently compute y.shape[1] sets of coefficients,
whereas setting it to true will compute one set of coefficients shared
across all the outputs. In that case, basisEval should be an array of
shape (n-by-N-by-D), with n the size of the dataset, N the number of
basis functions and D the number of outputs.
Parameters
----------
weights : list or numpy.ndarray, optional
The arrays for the weighted least-squares minimization. The default
is None.
Returns
-------
None.
'''
super().__init__(tensap.SquareLossFunction())
self.weights = weights
self.linear_solver = 'qr'
self.shared_coefficients = True
def __init__(self):
'''
Constructor for the class LinearModelLearningDensityL2.
Returns
-------
None.
'''
super().__init__(tensap.SquareLossFunction())
self.is_basis_orthonormal = True
range(DEGREE + 1))
for X_TRAIN in X.random_variables
]
BASES = tensap.FunctionalBases(BASES)
# %% Training and test samples
NUM_TRAIN = 1000
X_TRAIN = X.random(NUM_TRAIN)
Y_TRAIN = fun(X_TRAIN)
NUM_TEST = 10000
X_TEST = X.random(NUM_TEST)
Y_TEST = fun(X_TEST)
# %% Learning in canonical tensor format
SOLVER = tensap.CanonicalTensorLearning(ORDER, tensap.SquareLossFunction())
SOLVER.rank_adaptation = True
SOLVER.initialization_type = 'mean'
SOLVER.tolerance['on_error'] = 1e-6
SOLVER.alternating_minimization_parameters['stagnation'] = 1e-8
SOLVER.alternating_minimization_parameters['max_iterations'] = 100
SOLVER.linear_model_learning.regularization = False
SOLVER.linear_model_learning.basis_adaptation = True
SOLVER.bases = BASES
SOLVER.training_data = [X_TRAIN, Y_TRAIN]
SOLVER.display = True
SOLVER.alternating_minimization_parameters['display'] = False
SOLVER.test_error = True
SOLVER.test_data = [X_TEST, Y_TEST]
SOLVER.alternating_minimization_parameters['one_by_one_factor'] = False
SOLVER.alternating_minimization_parameters['inner_loops'] = 2
X_TEST = T.map(X_TEST)
# %% Tree-based tensor format
# Tensor format
# 1 - Random tree and active nodes
# 2 - Tensor-Train
# 3 - Hierarchial Tensor-Train
# 4 - Binary tree
CHOICE = 3
if CHOICE == 1:
print('Random tree with active nodes')
ARITY = [2, 4]
TREE = tensap.DimensionTree.random(ORDER, ARITY)
IS_ACTIVE_NODE = np.full(TREE.nb_nodes, True)
SOLVER = tensap.TreeBasedTensorLearning(TREE, IS_ACTIVE_NODE,
tensap.SquareLossFunction())
elif CHOICE == 2:
print('Tensor-train format')
SOLVER = tensap.TreeBasedTensorLearning.tensor_train(
ORDER, tensap.SquareLossFunction())
elif CHOICE == 3:
print('Tensor Train Tucker')
SOLVER = tensap.TreeBasedTensorLearning.tensor_train_tucker(
ORDER, tensap.SquareLossFunction())
elif CHOICE == 4:
print('Binary tree')
TREE = tensap.DimensionTree.balanced(ORDER)
IS_ACTIVE_NODE = np.full(TREE.nb_nodes, True)
SOLVER = tensap.TreeBasedTensorLearning(TREE, IS_ACTIVE_NODE,
tensap.SquareLossFunction())
else: