Principle:Kubeflow Pipelines Incremental Model Training
Domains: Machine_Learning, Training
Last Updated: 2026-02-13
Overview
A training strategy where additional boosting iterations are added to an existing model rather than training from scratch.
Description
Incremental (continued) training starts from a previously saved model and adds more boosting iterations. This allows iterative refinement — training a few rounds, evaluating, and training more if the model isnt good enough. The XGBoost starting_model parameter enables this by loading an existing model as the starting point.
Usage
Use in iterative training loops where a model is progressively refined until convergence criteria are met.
Theoretical Basis
Sequential model improvement via gradient boosting continuation. Each additional iteration fits a new tree to the residual errors of the existing ensemble.
| Concept | Description |
|---|---|
| Gradient Boosting | Ensemble method that sequentially adds weak learners to minimize a loss function |
| Continuation | New trees are added on top of the existing ensemble without retraining prior trees |
| Convergence Criteria | A threshold (e.g., MSE < threshold) that determines when to stop adding iterations |