Choose a set of states S that will be used to construct the training data. For each state it but be efficient to calculate the expectation value of X. For each state evaluate the exact expectation value with a classical computer and the noisy expectation value with a quantum computer. Generate training data set using both noisy and exact expectation values. Use this ansatz with the fitted parameters to predict noise-free observable of arbitrary quantum states.