def test_init(self):
obj1 = make_autoTensor([[-3,3,7],[4,5,9]])
obj2 = make_autoTensor([[-3,3],[4,5],[0,0]])
obj1.requires_grad = True
obj = MatMul(obj1,obj2)
assert obj.channels[0].autoVariable == obj1
def test_init(self):
obj1 = make_autoTensor([[-3,3],[4,5]])
obj2 = make_autoTensor([[-3,3],[4,5]])
obj1.requires_grad = True
obj = Divide(obj1,obj2)
assert obj.channels[0].autoVariable == obj1
def test_init(self):
y = make_autoTensor(torch.ones(5, 1))
y_pred = make_autoTensor(torch.rand(5, 1))
loss = SquareError(y_pred=y_pred, y_target=y)
assert loss.channels[0].autoVariable == y_pred
def test_der(self):
obj1 = make_autoTensor([[-3,3],[4,5]])
obj2 = make_autoTensor([[-3,3],[4,5]])
obj2.requires_grad = True
obj = Substract(obj1,obj2)
obj.backprop(make_autoTensor([[1,1],[1,1]]))
assert torch.sum(obj2.grad.value + make_autoTensor([[1,1],[1,1]]).value) == 0
def test_der(self):
obj1 = make_autoTensor([[-3,3],[4,5]])
obj2 = make_autoTensor([[-3,3],[4,5]])
obj1.requires_grad = True
obj = Multiply(obj1,obj2)
obj.backprop(make_autoTensor([[1,1],[1,1]]))
assert torch.sum(obj1.grad.value - obj2.value) == 0
def test_der(self):
obj1 = make_autoTensor([[-3,3],[4,5]])
obj1.requires_grad = True
obj = relu(obj1)
obj.backprop(make_autoTensor([[1,1],[1,1]]))
assert torch.sum(obj1.grad.value - make_autoTensor([[0,1],[1,1]]).value) == 0
def test_der(self):
obj1 = make_autoTensor([[-3,3,7],[4,5,9]])
obj2 = make_autoTensor([[-3,3],[4,5],[0,0]])
obj1.requires_grad = True
obj = MatMul(obj1,obj2)
obj.backprop(make_autoTensor([[1,1],[1,1]]))
assert torch.sum(obj1.grad.value - torch.mm(make_autoTensor([[1,1],[1,1]]).value,obj2.value.transpose(1,0))) == 0
def test_der(self):
obj1 = make_autoTensor([[-3,3],[4,5]])
obj1.requires_grad = True
obj = Power(obj1,3)
obj.backprop(make_autoTensor([[1,1],[1,1]]))
assert torch.sum(obj1.grad.value - 3 * make_autoTensor([[1,1],[1,1]]).value * obj1.value**2 ) == 0
def test_der(self):
obj1 = make_autoTensor([[-3,3],[4,5]])
obj1.requires_grad = True
obj = Sum(obj1)
obj.backprop(make_autoTensor([1,1]))
assert torch.sum(obj1.grad.value - make_autoTensor([1,1]).value * torch.ones(2,2) ) == 0
def test_der(self):
y = make_autoTensor(torch.ones(5, 1))
y_pred = make_autoTensor(torch.rand(5, 1))
y_pred.requires_grad = True
loss = SquareError(y_pred, y)
loss.backward()
assert torch.equal(y_pred.grad.value, y_pred.value - y.value)
def test_der(self):
obj1 = make_autoTensor([[-3,3],[4,5]])
obj2 = make_autoTensor([[-3,3],[4,5]])
obj1.requires_grad = True
obj2.requires_grad = True
obj = Divide(obj1,obj2)
obj.backprop(make_autoTensor([[1,1],[1,1]]))
assert torch.sum(obj1.grad.value - 1/obj1.value) == 0
assert torch.sum(obj2.grad.value + obj1.value/obj2.value**2) == 0
def test_make_autoTensor(self):
obj = make_autoTensor(0)
assert isinstance(obj, autoTensor)
obj = make_autoTensor(True)
assert isinstance(obj, autoTensor)
obj = make_autoTensor(torch.rand(1))
assert isinstance(obj, autoTensor)
obj = make_autoTensor(obj)
assert isinstance(obj, autoTensor)
def test_grad_sweep(self):
obj1 = make_autoTensor([[-3, 3], [4, 5]])
obj2 = make_autoTensor([[-3, 3], [4, 5]])
obj3 = make_autoTensor([[-3, 3], [4, 5]])
obj4 = make_autoTensor([[1, 1], [1, 1]])
obj1.requires_grad = True
obj2.requires_grad = False
obj3.requires_grad = True
obj4.requires_grad = True
obj5 = (obj1 + obj2) + (obj3 * obj4)
obj5.backprop(make_autoTensor([[-3, 3], [4, 5]]))
obj5.grad_sweep()
assert torch.sum(obj1.grad.value) == 0
assert obj2.grad == None
assert torch.sum(obj3.grad.value) == 0
assert torch.sum(obj4.grad.value) == 0
def test_grad_zeros(self):
obj = make_autoTensor([[-3, 3], [4, 5]])
obj.grad = make_autoTensor([[-3, 3], [4, 5]])
obj.grad_zeros()
assert torch.sum(obj.grad.value) == 0
