Classic machine learning

These examples run classic machine learning tasks on AI Runtime. They show how to use GPU acceleration for traditional ML algorithms, tabular foundation models, and time series forecasting, including XGBoost regression, zero-shot prediction with TabFM, and probabilistic forecasting with GluonTS.

Example	Interface	Description
XGBoost model training	Notebook	This notebook demonstrates how to train an XGBoost regression model on a single GPU. XGBoost can significantly benefit from GPU acceleration for large datasets.
Tune XGBoost hyperparameters with Optuna	Notebook	This notebook demonstrates hyperparameter optimization for an XGBoost classification model on a single GPU using Optuna and MLflow experiment tracking.
Zero-shot tabular predictions with TabFM	Notebook	This notebook demonstrates zero-shot classification and regression with Google's TabFM tabular foundation model. TabFM uses in-context learning to make predictions in a single forward pass, with no fine-tuning or dataset-specific training.
Time series forecasting with GluonTS	Notebook	This notebook demonstrates an end-to-end workflow for probabilistic time-series forecasting of electricity consumption data with GluonTS's DeepAR model on AI Runtime. It covers data ingestion, resampling, model training, prediction, visualization, and evaluation.
XGBoost training on a single GPU	CLI	Train an XGBoost classifier on the Forest CoverType dataset on one A10 GPU, log metrics to MLflow, and register the model in Unity Catalog.
XGBoost hyperparameter search with Ray Tune	CLI	Use Ray Tune and ASHA to search XGBoost hyperparameters across four A10 GPUs, log the best configuration to MLflow, and register the model in Unity Catalog.
Graph node classification with GraphSAGE	CLI	Train a GraphSAGE model on one A10 GPU to classify research papers in the ogbn-arxiv citation graph.
Time-series forecasting with Chronos-2	CLI	Use Chronos-2 to generate and evaluate rolling day-ahead electricity-price forecasts on one A10 GPU.
Time-series anomaly detection with Kumo-Anomaly	CLI	Fine-tune NVIDIA Kumo-Anomaly on one A10 GPU and compare the base and fine-tuned models on held-out anomaly-detection data.

Example	Interface	Description
XGBoost model training	Notebook	This notebook demonstrates how to train an XGBoost regression model on a single GPU. XGBoost can significantly benefit from GPU acceleration for large datasets.
Tune XGBoost hyperparameters with Optuna	Notebook	This notebook demonstrates hyperparameter optimization for an XGBoost classification model on a single GPU using Optuna and MLflow experiment tracking.
Zero-shot tabular predictions with TabFM	Notebook	This notebook demonstrates zero-shot classification and regression with Google's TabFM tabular foundation model. TabFM uses in-context learning to make predictions in a single forward pass, with no fine-tuning or dataset-specific training.
Time series forecasting with GluonTS	Notebook	This notebook demonstrates an end-to-end workflow for probabilistic time-series forecasting of electricity consumption data with GluonTS's DeepAR model on AI Runtime. It covers data ingestion, resampling, model training, prediction, visualization, and evaluation.
XGBoost training on a single GPU	CLI	Train an XGBoost classifier on the Forest CoverType dataset on one A10 GPU, log metrics to MLflow, and register the model in Unity Catalog.
XGBoost hyperparameter search with Ray Tune	CLI	Use Ray Tune and ASHA to search XGBoost hyperparameters across four A10 GPUs, log the best configuration to MLflow, and register the model in Unity Catalog.
Graph node classification with GraphSAGE	CLI	Train a GraphSAGE model on one A10 GPU to classify research papers in the ogbn-arxiv citation graph.
Time-series forecasting with Chronos-2	CLI	Use Chronos-2 to generate and evaluate rolling day-ahead electricity-price forecasts on one A10 GPU.
Time-series anomaly detection with Kumo-Anomaly	CLI	Fine-tune NVIDIA Kumo-Anomaly on one A10 GPU and compare the base and fine-tuned models on held-out anomaly-detection data.