Policy Gradient Methods for Discrete Time Linear Quadratic Regulator With Random Parameters