Implementation:Kubeflow Pipelines XGBoost Train Incremental

Sources: Kubeflow Pipelines, XGBoost

Domains: Machine_Learning, Training

Last Updated: 2026-02-13

Overview

Wrapper Doc for using the XGBoost training component with a starting_model for incremental training.

Description

This is the same xgboost_train_on_csv_op component used in XGBoost_Model_Training, but specifically for incremental training where starting_model is provided. The component loads the existing model and adds num_iterations more boosting rounds.

Code Reference

Source: samples/core/train_until_good/train_until_good.py (L36-42)

Import: from kfp import components

model = xgboost_train_on_csv_op(
    training_data=training_data,
    starting_model=starting_model,
    label_column=0,
    objective='reg:squarederror',
    num_iterations=50,
).outputs['model']

I/O Contract

Inputs

Name	Type	Required	Description
training_data	CSV	Yes	Training data in CSV format
starting_model	XGBoostModel	Yes	Previously trained model to continue from
label_column	int	Yes	Index of the label column
objective	str	Yes	XGBoost objective function (e.g., reg:squarederror)
num_iterations	int	Yes	Number of additional boosting rounds to add

Outputs

Name	Type	Description
model	XGBoostModel	Incrementally improved model with additional boosting rounds

Related Pages

• Principle:Kubeflow_Pipelines_Incremental_Model_Training