def test_compare_bp_and_rprop(self):
compare_networks(
# Test classes
partial(algorithms.GradientDescent, batch_size=None),
partial(algorithms.RPROP, maxstep=0.1),
# Test data
(simple_x_train, simple_y_train),
# Network configurations
network=self.network,
step=0.1,
shuffle_data=True,
verbose=False,
# Test configurations
epochs=50,
show_comparison_plot=False)
def test_compare_bp_and_hessian(self):
x_train, x_test, y_train, y_test = simple_classification()
compare_networks(
# Test classes
partial(algorithms.GradientDescent, batch_size=None),
partial(algorithms.Hessian, penalty_const=1),
# Test data
(x_train, y_train, x_test, y_test),
# Network configurations
network=[layers.Input(10),
layers.Sigmoid(15),
layers.Sigmoid(1)],
shuffle_data=True,
verbose=False,
show_epoch=1,
# Test configurations
epochs=5,
show_comparison_plot=False)
def test_compare_lvq_and_lvq21(self):
dataset = datasets.load_iris()
data, target = dataset.data, dataset.target
# Prepare the same weights for the fair comparison
lvq = algorithms.LVQ(n_inputs=4, n_subclasses=3, n_classes=3)
lvq.train(data, target, epochs=1)
prepared_lvq_weights = lvq.weight
compare_networks(
algorithms.LVQ,
partial(algorithms.LVQ21, epsilon=0.1),
data=[data, target],
epochs=10,
show_comparison_plot=False,
n_inputs=4,
n_subclasses=3,
n_classes=3,
weight=prepared_lvq_weights,
)
def test_compare_lvq_and_lvq3(self):
dataset = datasets.load_iris()
data, target = dataset.data, dataset.target
# Prepare the same weights for the fair comparison
lvq = algorithms.LVQ(n_inputs=4, n_subclasses=6, n_classes=3)
lvq.train(data, target, epochs=1)
prepared_lvq_weights = lvq.weight
compare_networks(
algorithms.LVQ,
partial(algorithms.LVQ3, epsilon=0.4),
data=[data, target],
epochs=100,
show_comparison_plot=False,
n_inputs=4,
n_subclasses=6,
n_classes=3,
prototypes_per_class=[4, 1, 1],
step=0.001,
weight=prepared_lvq_weights,
)
def test_momentum_with_minibatch(self):
x_train, _, y_train, _ = simple_classification()
compare_networks(
# Test classes
partial(algorithms.Momentum, batch_size=None),
partial(algorithms.Momentum, batch_size=1),
# Test data
(x_train, y_train),
# Network configurations
network=[
layers.Input(10),
layers.Sigmoid(20),
layers.Sigmoid(1)
],
step=0.25,
momentum=0.1,
shuffle_data=True,
verbose=False,
# Test configurations
epochs=40,
show_comparison_plot=False,
)
def test_compare_bp_and_hessian(self):
x_train, _, y_train, _ = simple_classification()
params = dict(weight=init.Uniform(-0.1, 0.1),
bias=init.Uniform(-0.1, 0.1))
compare_networks(
# Test classes
partial(algorithms.GradientDescent, batch_size=None),
partial(algorithms.HessianDiagonal, min_eigval=0.1),
# Test data
(x_train, y_train),
# Network configurations
network=[
layers.Input(10),
layers.Sigmoid(20, **params),
layers.Sigmoid(1, **params),
],
step=0.1,
shuffle_data=True,
verbose=False,
# Test configurations
epochs=50,
show_comparison_plot=False)
def test_compare_bp_and_cg(self):
x_train, x_test, y_train, y_test = simple_classification()
compare_networks(
# Test classes
partial(
partial(algorithms.GradientDescent, batch_size=None),
step=1.0,
),
partial(algorithms.ConjugateGradient,
update_function='fletcher_reeves'),
# Test data
(asfloat(x_train), asfloat(y_train)),
# Network configurations
network=layers.join(
layers.Input(10),
layers.Sigmoid(5),
layers.Sigmoid(1),
),
loss='mse',
shuffle_data=True,
# Test configurations
epochs=50,
show_comparison_plot=False)