Add HDFS save/load model support in XGBoost and LightGBM.

[lalalapotter]

Per customers requirement, we need to add support for saving/loading model on HDFS in XGBoost, and LightGBM.

Current workaround:

def save_model(model, path):
    if is_local_path(path):
        model.saveModel(path)
    else:
        file_name = str(uuid.uuid1())
        file_name = append_suffix(file_name, path)
        temp_path = os.path.join(tempfile.gettempdir(), file_name)
        try:
            model.saveModel(temp_path)
            put_local_file_to_remote(temp_path, path)
        finally:
            os.remove(temp_path)

def load_model(path, class_num):
    model = None
    if is_local_path(path):
        model = XGBClassifierModel.loadModel(path, class_num)
    else:
        file_name = str(uuid.uuid1())
        file_name = append_suffix(file_name, path)
        temp_path = os.path.join(tempfile.gettempdir(), file_name)
        get_remote_file_to_local(path, temp_path)
        try:
            model = XGBClassifierModel.loadModel(temp_path, class_num)
        finally:
            os.remove(temp_path)
    return model

eta = 0.1
max_depth = 6
num_round =200
params.update({"eta": eta, "max_depth": max_depth, "num_round": num_round})
classifier = XGBClassifier(params)
xgbmodel = classifier.fit(train)
save_model(xgbmodel, "hdfs:///user/kai/zcg/model/xgb.model")
xgbmodel = load_model("hdfs:///user/kai/zcg/model/xgb.model", 2)
xgbmodel.setFeaturesCol("features")
predicts = xgbmodel.transform(test).drop("features")
predicts.cache()
predicts.show(5, False)
predicts.unpersist(blocking=True)