If training set is linearly separable then Perceptron finds linear classifier separating data with 0 risk/loss eventually. But what if we believe data are separable but they
