def test_bad_n_components():
with pytest.raises(ValueError):
dcca = DCCA(3, 5, 5.5, [10, 10], [10, 10])
with pytest.raises(ValueError):
dcca = DCCA(3, 5, 0, [10, 10], [10, 10])
with pytest.raises(ValueError):
dcca = DCCA(10, 10, 4, [10, 2], [10, 3])
def test_bad_layer_sizes():
with pytest.raises(ValueError):
layer_sizes1, layer_sizes2 = [10, 0], [10, 10]
dcca = DCCA(3, 5, 2, layer_sizes1, layer_sizes2)
with pytest.raises(ValueError):
layer_sizes1, layer_sizes2 = [10, 5.4], [10, 10]
dcca = DCCA(3, 5, 2, layer_sizes1, layer_sizes2)
with pytest.raises(ValueError):
layer_sizes1, layer_sizes2 = [10, 5], [0, 10]
dcca = DCCA(3, 5, 2, layer_sizes1, layer_sizes2)
with pytest.raises(ValueError):
layer_sizes1, layer_sizes2 = [10, 5], [1.9, 10]
dcca = DCCA(3, 5, 2, layer_sizes1, layer_sizes2)
with pytest.raises(ValueError):
layer_sizes1, layer_sizes2 = [10, 5], np.array([10, 10])
dcca = DCCA(3, 5, 2, layer_sizes1, layer_sizes2)
with pytest.raises(ValueError):
layer_sizes1, layer_sizes2 = np.array([10, 10]), [10, 10]
dcca = DCCA(3, 5, 2, layer_sizes1, layer_sizes2)
with pytest.raises(ValueError):
layer_sizes1, layer_sizes2 = (10, 5, 4)
dcca = DCCA(3, 5, 2, layer_sizes1, layer_sizes2)
with pytest.raises(ValueError):
layer_sizes1, layer_sizes2 = [10, 5], [10, 10]
dcca = DCCA(3, 5, 2, layer_sizes1, layer_sizes2)
def data():
random_seed = 10
N = 600
input_size1, input_size2 = 100, 100
sin_transform = []
np.random.seed(random_seed)
sin_transform.append(20 * np.random.rand(N, input_size1))
sin_transform.append(np.sin(sin_transform[0]))
layer_sizes1 = [1024, 50]
layer_sizes2 = [1024, 50]
dcca_50_2 = DCCA(input_size1,
input_size2,
n_components=2,
layer_sizes1=[1024, 50],
layer_sizes2=[1024, 50])
dcca_print = DCCA(input_size1,
input_size2,
n_components=2,
layer_sizes1=[1024, 50],
layer_sizes2=[1024, 50],
print_train_log_info=True)
dcca_low_epochs = DCCA(input_size1,
input_size2,
n_components=2,
layer_sizes1=[1024, 80],
layer_sizes2=[1024, 80],
epoch_num=10)
dcca_use_all = DCCA(input_size1,
input_size2,
n_components=2,
layer_sizes1=[1024, 80],
layer_sizes2=[1024, 80],
use_all_singular_values=True)
dcca_1layer = DCCA(input_size1, input_size2, n_components=2)
return {
'N': N,
'input_size1': input_size1,
'input_size2': input_size2,
'layer_sizes1': layer_sizes1,
'layer_sizes2': layer_sizes2,
'sin_transform': sin_transform,
'random_seed': random_seed,
'dcca_50_2': dcca_50_2,
'dcca_low_epochs': dcca_low_epochs,
'dcca_print': dcca_print,
'dcca_use_all': dcca_use_all,
'dcca_1layer': dcca_1layer
}
def test_bad_tolerance(data):
with pytest.raises(ValueError):
dcca = DCCA(5,
5,
2,
data['layer_sizes1'],
data['layer_sizes2'],
tolerance=0)
with pytest.raises(ValueError):
dcca = DCCA(5,
5,
2,
data['layer_sizes1'],
data['layer_sizes2'],
tolerance=-4)
def test_bad_reg_par(data):
with pytest.raises(ValueError):
dcca = DCCA(5,
5,
2,
data['layer_sizes1'],
data['layer_sizes2'],
reg_par=0)
with pytest.raises(ValueError):
dcca = DCCA(5,
5,
2,
data['layer_sizes1'],
data['layer_sizes2'],
reg_par=-4)
def test_bad_learning_rate(data):
with pytest.raises(ValueError):
dcca = DCCA(5,
5,
2,
data['layer_sizes1'],
data['layer_sizes2'],
learning_rate=0)
with pytest.raises(ValueError):
dcca = DCCA(5,
5,
2,
data['layer_sizes1'],
data['layer_sizes2'],
learning_rate=-4)
def test_bad_batch_size(data):
with pytest.raises(ValueError):
dcca = DCCA(5,
5,
2,
data['layer_sizes1'],
data['layer_sizes2'],
batch_size=0)
with pytest.raises(ValueError):
dcca = DCCA(5,
5,
2,
data['layer_sizes1'],
data['layer_sizes2'],
batch_size=91.9)
def test_bad_epoch_num(data):
with pytest.raises(ValueError):
dcca = DCCA(5,
5,
2,
data['layer_sizes1'],
data['layer_sizes2'],
epoch_num=0)
with pytest.raises(ValueError):
dcca = DCCA(5,
5,
2,
data['layer_sizes1'],
data['layer_sizes2'],
epoch_num=91.9)
def test_bad_input_size(data):
# incorrect size
with pytest.raises(ValueError):
bad_data1 = []
bad_data1.append(np.random.rand(50, 10))
bad_data1.append(np.random.rand(50, data['input_size2']))
#dcca_50_2.fit(bad_data1)
data['dcca_50_2'].fit(bad_data1)
with pytest.raises(ValueError):
bad_data1 = []
bad_data1.append(np.random.rand(50, data['input_size1']))
bad_data1.append(np.random.rand(50, 5))
# dcca_50_2.fit(bad_data1)
data['dcca_50_2'].fit(bad_data1)
with pytest.raises(ValueError):
dcca = DCCA(0, 5, 2, [10, 10], [10, 10])
with pytest.raises(ValueError):
dcca = DCCA(2, 0, 2, [10, 10], [10, 10])
with pytest.raises(ValueError):
dcca = DCCA(0.6, 5, 2, [10, 10], [10, 10])
with pytest.raises(ValueError):
dcca = DCCA(2, 10.1, 2, [10, 10], [10, 10])
['Raw Views', 'CCA', 'Polynomial KCCA', 'Gaussian KCCA', 'DCCA']
transform_labels = \
['Linear Transform', 'Polynomial Transform', 'Sinusoidal Transform']
input_size1 = Xs_train_sets[0][0].shape[1]
input_size2 = Xs_train_sets[0][1].shape[1]
outdim_size = min(Xs_train_sets[0][0].shape[1], 2)
layer_sizes1 = [256, 256, outdim_size]
layer_sizes2 = [256, 256, outdim_size]
methods = [
CCA(regs=0.1, n_components=2),
KMCCA(kernel='poly', regs=0.1, kernel_params={'degree': 2, 'coef0': 0.1},
n_components=2),
KMCCA(kernel='rbf', regs=0.1, kernel_params={'gamma': 1/4},
n_components=2),
DCCA(input_size1, input_size2, outdim_size, layer_sizes1, layer_sizes2,
epoch_num=400)
]
fig, axes = plt.subplots(3 * 2, 5 * 2, figsize=(22, 12))
sns.set_context('notebook')
for r, transform in enumerate(transforms):
axs = axes[2 * r:2 * r + 2, :2]
for i, ax in enumerate(axs.flatten()):
dim2 = int(i / 2)
dim1 = i % 2
ax.scatter(
Xs_test_sets[r][0][:, dim1],
Xs_test_sets[r][1][:, dim2],
cmap=cmap,
c=labels,
###############################################################################
# Fit DCCA model to uncover latent distribution
# ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
#
# The output dimensionality is still 4.
# Define parameters and layers for deep model
features1 = Xs_train[0].shape[1] # Feature sizes
features2 = Xs_train[1].shape[1]
layers1 = [256, 256, 4] # nodes in each hidden layer and the output size
layers2 = [256, 256, 4]
dcca = DCCA(input_size1=features1,
input_size2=features2,
n_components=4,
layer_sizes1=layers1,
layer_sizes2=layers2,
epoch_num=500)
dcca.fit(Xs_train)
Xs_transformed = dcca.transform(Xs_test)
###############################################################################
# Visualize the transformed data
# ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
#
# We can see that it has uncovered the latent correlation between views.
crossviews_plot(Xs_transformed,
labels=y_test,
title='Transformed Testing Data View 1 vs. View 2 '
'(Polynomial Transform + noise)',