def test_variable_sum():
a_array = np.arange(5)
a = Variable(np.arange(5, dtype=float))
expected_result_variable = Variable(10)
expected_result_number = 10
assert np.all(a_array == a.data)
assert np.all(expected_result_variable == expected_result_number)
assert id(a_array) != id(a.data)
assert id(expected_result_variable) != id(expected_result_number)
def validate_variable_result(result) -> None:
assert result.eq(expected_result_variable).all()
assert isinstance(result, Variable)
return
def validate_number_result(result) -> None:
assert np.all(result == expected_result_number)
float(result) # error means it can't be converted to a float
assert isinstance(result, np.number)
return
# Variable
validate_variable_result(a.sum())
validate_variable_result(np.sum(a))
# numpy
validate_number_result(a_array.sum())
validate_number_result(np.sum(a_array))
# Verify Derivative
sgd = autograd.optimizer.SGD(learning_rate=1e-3)
summation = a.sum()
variable_to_gradient = sgd.take_training_step(summation)
assert all(
isinstance(var, Variable) and isinstance(grad, np.ndarray)
for var, grad in variable_to_gradient.items())
assert np.all(variable_to_gradient[a] == np.ones(a.shape))
# Verify Trainability
x = Variable(np.random.rand(2))
y = 10
sgd = autograd.optimizer.SGD(learning_rate=1e-1)
for training_step_index in range(1_000):
y_hat = x.sum()
diff = np.subtract(y, y_hat)
loss = diff**2
if training_step_index > 10 and loss.sum() < 1e-10:
break
sgd.take_training_step(loss)
isinstance(var, Variable) and isinstance(grad, np.ndarray)
for var, grad in variable_to_gradient.items())
assert np.all(variable_to_gradient[a] == np.ones(a.shape))
# Verify Trainability
x = Variable(np.random.rand(2))
y = 10
sgd = autograd.optimizer.SGD(learning_rate=1e-1)
for training_step_index in range(1_000):
y_hat = x.sum()
diff = np.subtract(y, y_hat)
loss = diff**2
if training_step_index > 10 and loss.sum() < 1e-10:
break
sgd.take_training_step(loss)
assert np.abs(x.sum() - 10) < 1e-4
assert loss.sum() < 1e-10
def test_variable_abs():
a_array = np.array([0, -1, -2, 3])
a = Variable(np.array([0, -1, -2, 3], dtype=float))
expected_result_variable = Variable(np.arange(4))
expected_result_array = np.arange(4)
assert np.all(a_array == a.data)
assert np.all(expected_result_variable == expected_result_array)
assert id(a_array) != id(a.data)
assert id(expected_result_variable) != id(expected_result_array)
