def compute_cost(X, Y, W, b, lambd):
    P, H = evaluate_classifier2(X, W, b)
    n = X.shape[1]

    loss = 0
    for i in range(n):
        loss -= np.log( Y[:, i:i+1].T @ P[:, i:i+1] )
    loss = np.asscalar(loss) / n

    regularization = 0
    for i in range(np.shape(W)[0]):
        regularization += lambd * np.sum(W[i]**2)

    J = loss + regularization
    return J#, loss

def compute_cost_old(X,Y,W,b,lambd,return_loss=False):
    p = evaluate_classifier(X,W,b)
    reg_term =  lambd * np.sum( [np.sum(np.square(W_i)) for W_i in W] )
    loss = 0
    #print('\ncomputing cost:\nP shape: {}'.format(p.shape))
    for i in range(X.shape[1]):
            loss += - np.log(Y[:,i:i+1].T @ p[:,i:i+1])
    loss = np.asscalar(loss)
    loss /= X.shape[1]
    cost = loss + reg_term
    #data_size = X.shape[1] if len(X.shape) == 2 else 1

    if return_loss:
            return cost.item(0), loss.item(0)
    return cost.item(0)