def test_forward_pass(self):
# Really just testing that the forward pass doesn't blow up
mlp = DistributedMLP(10, (3, 3), 2)
x = torch.randn(1, 10)
result = mlp.forward(x)
print(f'The result of the forward pass is {result}')
assert result is not None
def test_test_model(self):
mlp = DistributedMLP(2, (3, 3), 1)
optimizer = SGD(mlp.parameters(), lr=0.001, momentum=0.9)
loss_function = torch.nn.L1Loss()
dataset = TensorDataset(self.X, self.y_1)
data_loader = DataLoader(dataset)
test_dataset = TensorDataset(self.X_test, self.y_test_1)
test_data_loader = DataLoader(test_dataset)
train(mlp, 100, data_loader, optimizer, loss_function)
test_loss, accuracy = test(mlp, test_data_loader, loss_function)
assert test_loss is not None
assert accuracy is not None
def test_train_model_to_predict_1(self):
mlp = DistributedMLP(2, (3, 3), 1)
dataset = TensorDataset(self.X, self.y_1) # create your datset
batch_size = 3
data_loader = DataLoader(dataset, batch_size)
optimizer = SGD(mlp.parameters(), lr=0.001, momentum=0.9)
loss_function = torch.nn.L1Loss()
train(mlp, 100, data_loader, optimizer, loss_function)
y_pred = mlp(self.X_test)
assert y_pred is not None
for y in y_pred:
assert 1 - y < 1e-2
def test_weights_are_initialised_with_std_kaiming_init(self):
mlp = DistributedMLP(10, (5, 5), 2)
for module in list(mlp._modules.items()):
standard_deviation = module[1].weight.std()
expected = self.calc_expected_standard_deviation(module[1].weight.size(1))
print(f'Actual Standard deviation {standard_deviation} | Expected {expected} ')
assert abs(standard_deviation - expected) < 1e4
def test_weights_are_initialised_with_mean_close_to_zero(self):
mlp = DistributedMLP(10, (5, 5), 2)
for module in list(mlp._modules.items()):
mean = module[1].weight.mean()
print(f'The mean is {mean}')
assert module[1].weight.mean().abs() < 12e-2
def test_model_is_created_correctly(self):
mlp = DistributedMLP(10, (5, 5), 2)
assert mlp is not None