3. Suppose we have a prediction problem where the target t corresponds to an angle, measured in radians. A reasonable loss function for this can be L(y,t)=1cos(yt), where y is the prediction for the input x. Suppose we make the prediction by using a linear model y=wTx+b. a. Derive a sequence of expressions for the gradient of cost with respect to y,w and b(yL,wL and bL) b. Assume that we have N training examples in a design matrix X (each row of X is a training instance). Then the overall costE=N1i=1nL(yi,ti). Derive a sequence of vectorized expressions for the gradient of costE with respect to y, w and b, i.e., yE,wE and bE. Your expression should be such that you would be able to use those for a for-loop free implementation in Python..