5. Building the Tensor

1. The Wrapper

We wrap NumPy arrays in a Tensor class that tracks gradients.

python
class Tensor:
    def __init__(self, data, requires_grad=False):
        self.data = np.array(data)
        self.grad = None
        self.requires_grad = requires_grad
        self._backward = lambda: None

2. Dependency Tracking

When we add two Tensors, the result must remember its parents to backpropagate later. This forms a Directed Acyclic Graph (DAG).