How to use continuation learning for XGBoost with MultiOutputRegressor?

Question

I'm trying to train multi-output XGBoost regression model using the continuation training, but I get this error:

TypeError: ('Unknown type:', MultiOutputRegressor(estimator=XGBRegressor(base_score=None, booster=None)

My code is as follows:

params = {
    'n_estimators':2000,
    'learning_rate':0.001,
    'max_depth':6,
    'min_child_weight': 1,
    'eta':.3,
    'subsample': 1,
    'colsample_bytree': 1,
    'objective':'reg:squarederror',}
    path = os.path.join('.', "xgb_model_direct.pkl")

for i, (x_batch, y_batch, _, _, _, _) in enumerate(pca_train_ds):

    # transform X: (batch_size, num_components, input_horizon_steps) --> 
    # (batch_size,  num_components*input_horizon_steps) 
    x_batch = x_batch.numpy().reshape(x_batch.shape[0], -1)
    
    # transform Y: (batch_size, num_components) --> 
    #(batch_size , num_components) 
    y_batch = y_batch.numpy().reshape(y_batch.shape[0], -1)

    # define model
    model = xgb.XGBRegressor(**params)
    
    # define the direct multioutput wrapper model
    wrapper = MultiOutputRegressor(model)

 if i==0:
    wrapper.fit(x_batch, y_batch)
    # save model to file
    joblib.dump(wrapper, path)
else:
    # load model from file
    loaded_model = joblib.load(path)
    wrapper.fit(x_batch, y_batch, xgb_model=loaded_model)