MPINestedGridSearchCV.py 文件源码

def _fit_and_score_with_parameters(X, y, cv, best_parameters):
    """Distribute work of non-nested cross-validation across slave nodes."""
    # tell slaves testing phase is next
    _task_desc = numpy.empty(2, dtype=int)
    _task_desc[1] = MPI_MSG_TEST

    comm.Bcast([_task_desc, MPI.INT], root=0)
    comm.bcast((X, y), root=0)

    # Compability with sklearn > 0.18 TODO
    _splitted_cv = [(a, b) for a, b in cv.split(X, y)]

    assert comm_size >= len(_splitted_cv)

    for i, (train_index, test_index) in enumerate(_splitted_cv):
        fold_id = i + 1
        LOG.info("Testing fold %d", fold_id)

        parameters = best_parameters.loc[fold_id, :].to_dict()
        work_item = (fold_id, train_index, test_index, parameters)

        comm.send(work_item, dest=fold_id, tag=MPI_TAG_TRAIN_TEST_DATA)

    scores = {}
    for i in range(len(_splitted_cv)):
        fold_id, test_result = comm.recv(source=MPI.ANY_SOURCE,
                                         tag=MPI_TAG_RESULT)
        scores[fold_id] = test_result

    # Tell all nodes to terminate
    for i in range(len(_splitted_cv), comm_size):
        comm.send((0, None), dest=i, tag=MPI_TAG_TRAIN_TEST_DATA)

    return pandas.Series(scores)