why do we assume the sample space to be the cartesian product of input and output space?

69 Views Asked by At

Why -- in supervised learning literature -- do we assume the sample space to be the cartesian product of input space i.e., $\mathcal{Z}=\mathcal{X}\times\mathcal{Y}$. Since there is a one-to one mapping in most cases. Why can't we consider $\mathcal{Z}=\{(x,y=g(x)): x\in\mathcal{X}\}$ where $g$ is some data generating function. It reduces the sample space size considerably. What I want to say is that is there a specific reason why we make the cartesian product supposition?