Using test_autoaugment_policy to calcualte the accuracy of subpolicies. ES learner now outputs similar list to randomsearcher