>>> from dl_numpy import Tensor, math
>>> a, b = Tensor([2, 1]), Tensor([-0.1, 0.3])
>>> out = math.sum(a * b)  # calculation
>>> grad = out.backward()  # backward
>>> grad[id(a)]  # gradient for a
[0.1 0.3]

>>> out = a * b + math.matmul(a, b) ** 2 / a
>>> out.graph.render('calc_graph', format='svg')  # save graph as a svg file

>>> from dl_numpy.layer.activation import sigmoid
>>> from dl_numpy.layer.loss import mse
>>> from dl_numpy.layer import Linear
>>> m = Linear(8, 32, activation=sigmoid)  # linear layer (Wx+b)
>>> out = m(x)  # forward
>>> loss = mse(y, out)  # MSE loss
>>> grad = loss.backward()  # backward
>>> m.apply_grad(grad, lr=0.01)  # update parameters