def test_mse_tensor():
true = torch.Tensor([11, 11])
pred = torch.Tensor([10, 10])
loss = mean_squared_error(true, pred)
assert loss == 1
true = torch.Tensor([10, 10])
pred = torch.Tensor([10, 10])
loss = mean_squared_error(true, pred)
assert loss == 0
def test_adam():
X = Variable(torch.rand(5, 3))
y = Variable(torch.rand(5))
w = Parameter(torch.rand(3))
opt = Adam(params=[w])
o = X @ w
loss = mean_squared_error(y, o)
loss.backward()
opt.step()
assert loss.data[0] > mean_squared_error(y, X @ w).data[0]
def test_sgd_momentum():
X = Variable(torch.rand(5, 3))
y = Variable(torch.rand(5))
w = Parameter(torch.rand(3))
opt = SGD(lr=0.1, momentum=0.99, params=[w])
o = X @ w
loss = mean_squared_error(y, o)
loss.backward()
opt.step()
assert loss.data[0] > mean_squared_error(y, X @ w).data[0]
def test_sgd_learning_rate():
X = Variable(torch.rand(5, 3))
y = Variable(torch.rand(5))
w = Variable(torch.rand(3), requires_grad=True)
opt = SGD(lr=0.1, params=[w])
o = X @ w
loss = mean_squared_error(y, o)
loss.backward()
opt.step()
assert loss > mean_squared_error(y, X @ w)
def test_mse_variable():
true = Variable(torch.Tensor([11, 11]))
pred = Variable(torch.Tensor([10, 10]))
loss = mean_squared_error(true, pred)
assert type(loss) is Variable
assert loss == 1