Your implementation is a little bit different from the formula in the paper, for example, where the denominator has sigma squared and yours doesn't 