How to map the classical data set to the Hamiltonian

VQA is used to find the minimum value of the Hamiltonian to ensure the minimum energy, so it is a parameterized quantum circuit training for the Hamiltonian. And our general QNN neural network training starts from the classical data set. The general method is to program the classical data into a quantum state, so what if I want to use VQA to train the classical data set? How can a classical data set be transformed into a Hamiltonian instead of a quantum state?