def test_conv_rbm_W():
new_conv_rbm = conv_rbm.ConvRBM()
assert new_conv_rbm.W.size(0) == 5
assert new_conv_rbm.W.size(1) == 1
assert new_conv_rbm.W.size(2) == 7
assert new_conv_rbm.W.size(3) == 7
def test_conv_rbm_reconstruct():
test = torchvision.datasets.KMNIST(
root="./data",
train=False,
download=True,
transform=torchvision.transforms.ToTensor(),
)
new_conv_rbm = conv_rbm.ConvRBM(
visible_shape=(28, 28),
filter_shape=(1, 1),
n_filters=1,
n_channels=1,
steps=1,
learning_rate=0.01,
momentum=0,
decay=0,
use_gpu=True,
)
e, v = new_conv_rbm.reconstruct(test)
assert e >= 0
assert v.size(1) == 1
assert v.size(2) == 28
assert v.size(3) == 28
def test_conv_rbm_energy():
new_conv_rbm = conv_rbm.ConvRBM()
samples = torch.ones(5, 1, 7, 7)
energy = new_conv_rbm.energy(samples)
assert torch.mean(energy).detach().numpy() < 0
def test_conv_rbm_hidden_shape_setter():
new_conv_rbm = conv_rbm.ConvRBM()
try:
new_conv_rbm.hidden_shape = 'a'
except:
new_conv_rbm.hidden_shape = (22, 22)
assert new_conv_rbm.hidden_shape == (22, 22)
def test_conv_rbm_b_setter():
new_conv_rbm = conv_rbm.ConvRBM()
try:
new_conv_rbm.b = 1
except:
new_conv_rbm.b = torch.nn.Parameter(torch.zeros(5))
assert new_conv_rbm.b.size(0) == 5
def test_conv_rbm_visible_shape_setter():
new_conv_rbm = conv_rbm.ConvRBM()
try:
new_conv_rbm.visible_shape = 'a'
except:
new_conv_rbm.visible_shape = (28, 28)
assert new_conv_rbm.visible_shape == (28, 28)
def test_conv_rbm_a_setter():
new_conv_rbm = conv_rbm.ConvRBM()
try:
new_conv_rbm.a = 1
except:
new_conv_rbm.a = torch.nn.Parameter(torch.zeros(1))
assert new_conv_rbm.a.size(0) == 1
def test_conv_rbm_visible_sampling():
new_conv_rbm = conv_rbm.ConvRBM()
h = torch.ones(1, 5, 7, 7)
probs, states = new_conv_rbm.visible_sampling(h)
assert probs.size(1) == 1
assert states.size(1) == 1
def test_conv_rbm_hidden_sampling():
new_conv_rbm = conv_rbm.ConvRBM()
v = torch.ones(5, 1, 7, 7)
probs, states = new_conv_rbm.hidden_sampling(v)
assert probs.size(1) == 5
assert states.size(1) == 5
def test_conv_rbm_filter_shape_setter():
new_conv_rbm = conv_rbm.ConvRBM()
try:
new_conv_rbm.filter_shape = "a"
except:
new_conv_rbm.filter_shape = (7, 7)
assert new_conv_rbm.filter_shape == (7, 7)
def test_conv_rbm_optimizer_setter():
new_conv_rbm = conv_rbm.ConvRBM()
try:
new_conv_rbm.optimizer = 'OPT'
except:
new_conv_rbm.optimizer = torch.optim.SGD(
new_conv_rbm.parameters(), lr=0.1)
assert type(new_conv_rbm.optimizer).__name__ == 'SGD'
def test_conv_rbm_fit():
train = torchvision.datasets.KMNIST(
root='./data', train=True, download=True, transform=torchvision.transforms.ToTensor())
new_conv_rbm = conv_rbm.ConvRBM(visible_shape=(28, 28), filter_shape=(1, 1), n_filters=1, n_channels=1,
steps=1, learning_rate=0.01, momentum=0, decay=0, use_gpu=True)
e = new_conv_rbm.fit(train, batch_size=128, epochs=1)
assert e >= 0
def test_conv_rbm_forward():
new_conv_rbm = conv_rbm.ConvRBM()
v = torch.ones(1, 1, 28, 28)
probs = new_conv_rbm.forward(v)
assert probs.size(1) == 5
assert probs.size(2) == 22
assert probs.size(3) == 22
def test_conv_rbm_W_setter():
new_conv_rbm = conv_rbm.ConvRBM()
try:
new_conv_rbm.W = 1
except:
new_conv_rbm.W = torch.nn.Parameter(torch.randn(5, 1, 7, 7) * 0.01)
assert new_conv_rbm.W.size(0) == 5
assert new_conv_rbm.W.size(1) == 1
assert new_conv_rbm.W.size(2) == 7
assert new_conv_rbm.W.size(3) == 7
def test_conv_rbm_steps_setter():
new_conv_rbm = conv_rbm.ConvRBM()
try:
new_conv_rbm.steps = "a"
except:
new_conv_rbm.steps = 1
assert new_conv_rbm.steps == 1
try:
new_conv_rbm.steps = 0
except:
new_conv_rbm.steps = 1
assert new_conv_rbm.steps == 1
def test_conv_rbm_decay_setter():
new_conv_rbm = conv_rbm.ConvRBM()
try:
new_conv_rbm.decay = "a"
except:
new_conv_rbm.decay = 0.1
assert new_conv_rbm.decay == 0.1
try:
new_conv_rbm.decay = -1
except:
new_conv_rbm.decay = 0.1
assert new_conv_rbm.decay == 0.1
def test_conv_rbm_lr_setter():
new_conv_rbm = conv_rbm.ConvRBM()
try:
new_conv_rbm.lr = "a"
except:
new_conv_rbm.lr = 0.1
assert new_conv_rbm.lr == 0.1
try:
new_conv_rbm.lr = -1
except:
new_conv_rbm.lr = 0.1
assert new_conv_rbm.lr == 0.1
def test_conv_rbm_n_channels_setter():
new_conv_rbm = conv_rbm.ConvRBM()
try:
new_conv_rbm.n_channels = "a"
except:
new_conv_rbm.n_channels = 1
assert new_conv_rbm.n_channels == 1
try:
new_conv_rbm.n_channels = 0
except:
new_conv_rbm.n_channels = 1
assert new_conv_rbm.n_channels == 1
def test_conv_rbm_n_filters_setter():
new_conv_rbm = conv_rbm.ConvRBM()
try:
new_conv_rbm.n_filters = 'a'
except:
new_conv_rbm.n_filters = 1
assert new_conv_rbm.n_filters == 1
try:
new_conv_rbm.n_filters = 0
except:
new_conv_rbm.n_filters = 1
assert new_conv_rbm.n_filters == 1
def test_conv_rbm_momentum_setter():
new_conv_rbm = conv_rbm.ConvRBM()
try:
new_conv_rbm.momentum = "a"
except:
new_conv_rbm.momentum = 0.1
assert new_conv_rbm.momentum == 0.1
try:
new_conv_rbm.momentum = -1
except:
new_conv_rbm.momentum = 0.1
assert new_conv_rbm.momentum == 0.1
def test_conv_rbm_steps():
new_conv_rbm = conv_rbm.ConvRBM()
assert new_conv_rbm.steps == 1
def test_conv_rbm_n_channels():
new_conv_rbm = conv_rbm.ConvRBM()
assert new_conv_rbm.n_channels == 1
def test_conv_rbm_visible_shape():
new_conv_rbm = conv_rbm.ConvRBM()
assert new_conv_rbm.visible_shape == (28, 28)
def test_conv_rbm_n_filters():
new_conv_rbm = conv_rbm.ConvRBM()
assert new_conv_rbm.n_filters == 5
def test_conv_rbm_hidden_shape():
new_conv_rbm = conv_rbm.ConvRBM()
assert new_conv_rbm.hidden_shape == (22, 22)
def test_conv_rbm_filter_shape():
new_conv_rbm = conv_rbm.ConvRBM()
assert new_conv_rbm.filter_shape == (7, 7)
def test_conv_rbm_a():
new_conv_rbm = conv_rbm.ConvRBM()
assert new_conv_rbm.a.size(0) == 1
def test_conv_rbm_b():
new_conv_rbm = conv_rbm.ConvRBM()
assert new_conv_rbm.b.size(0) == 5
def test_conv_rbm_optimizer():
new_conv_rbm = conv_rbm.ConvRBM()
assert type(new_conv_rbm.optimizer).__name__ == "SGD"
def test_conv_rbm_optimizer_setter():
new_conv_rbm = conv_rbm.ConvRBM()
new_conv_rbm.optimizer = torch.optim.SGD(new_conv_rbm.parameters(), lr=0.1)
assert type(new_conv_rbm.optimizer).__name__ == "SGD"