QP Optimiser

qp_opt.py

class QPoptimizer(object):
    def __call__(self, feature_num, learner, expert):
        w = cp.Variable(feature_num)

        obj_func = cp.Minimize(cp.norm(w))
        constraints = [(expert-learner) @ w >= 2]

        prob = cp.Problem(obj_func, constraints)
        prob.solve()

        if prob.status == "optimal":
            log.debug("status: {0}".format(prob.status))
            log.debug("optimal value {0}".format(prob.value))

            weights = np.squeeze(np.asarray(w.value))
            norm = np.linalg.norm(weights)
            weights = weights/norm
            return weights, prob.value
        else:
            log.debug("status: {0}".format(prob.status))
            log.debug("returning Zeros..")

            weights = np.zeros(feature_num)
            return weights, prob.status