def test_forward_multi_input(self, multi_input_dnn):
"""Test Forward of Multi Input ConvNet."""
master_device_setter(multi_input_dnn, 'cpu')
input_tensor = [torch.rand(size=[10, 1, 28]),
torch.rand(size=[10, 1, 28, 28])]
out = multi_input_dnn(input_tensor)
assert out.shape == (10, 50)
def test_fail_mixed_devices(self, multi_input_cnn, conv3D_net,
multi_input_dnn, conv1D_net,
multi_input_dnn_data, multi_input_cnn_data):
"""Test training throws ValueError when network has mixed devices."""
assert hasattr(conv1D_net, 'device')
assert hasattr(conv3D_net, 'device')
assert hasattr(multi_input_dnn, 'device')
assert hasattr(multi_input_cnn, 'device')
master_device_setter(multi_input_cnn, 'cuda:0')
assert conv3D_net == multi_input_cnn.input_networks['conv3D_net']
assert multi_input_dnn == multi_input_cnn.input_networks[
'multi_input_dnn']
data_len = len(multi_input_cnn_data)
train_loader = DataLoader(
Subset(multi_input_cnn_data, range(data_len // 2)))
valid_loader = DataLoader(
Subset(multi_input_cnn_data, range(data_len // 2, data_len)))
multi_input_cnn.fit(train_loader=train_loader,
val_loader=valid_loader,
epochs=1,
plot=False)
with pytest.raises(ValueError) as e_info:
multi_input_cnn.input_networks['conv3D_net'].device = 'cpu'
multi_input_cnn.fit(train_loader=train_loader,
val_loader=valid_loader,
epochs=1,
plot=False)
assert str(e_info.value).endswith("{'conv3D_net': device(type='cpu')}")
def test_master_device_setter(multi_input_cnn):
"""If CUDA is available, test master_device_setter usage."""
# Make sure the network is in cpu first
assert str(multi_input_cnn.device) == 'cpu'
master_device_setter(multi_input_cnn, device='cuda:0')
assert str(multi_input_cnn.device) == 'cuda:0'
assert str(list(multi_input_cnn.input_networks.values())[0] == 'cuda:0')
assert str(
list(
list(multi_input_cnn.input_networks.values())
[2].input_networks.values())[0] == 'cuda:0')
def test_master_net_device_set_to_cuda(self, multi_input_cnn):
"""Test if the network as whole gets switched to cuda."""
assert hasattr(multi_input_cnn, 'device')
master_device_setter(multi_input_cnn, 'cuda:0')
assert multi_input_cnn.device == torch.device(type='cuda', index=0)
assert multi_input_cnn.input_networks['conv3D_net']\
.device == torch.device(type='cuda', index=0)
assert multi_input_cnn.input_networks['multi_input_dnn']\
.device == torch.device(type='cuda', index=0)
assert multi_input_cnn.input_networks['multi_input_dnn'].\
input_networks['conv1D_net'].\
device == torch.device(type='cuda', index=0)
assert multi_input_cnn.input_networks['multi_input_dnn'].\
input_networks['conv2D_net'].\
device == torch.device(type='cuda', index=0)