This feature is subject to the "Pre-GA Offerings Terms" in the General Service Terms section of theService Specific Terms. Pre-GA features are available "as is" and might have limited support. For more information, see thelaunch stage descriptions.

This document provides an overview of Tabular Workflow for Forecastingpipeline and components. To learn how to train a model, seeTrain a model with Tabular Workflow for Forecasting.

Tabular Workflow for Forecasting is the complete pipeline forforecasting tasks. It is similar to theAutoML API,but lets you to choose what to control and what to automate. Instead of havingcontrols for thewhole pipeline, you have controls forevery step in thepipeline. These pipeline controls include:

• Data splitting
• Feature engineering
• Architecture search
• Model training
• Model ensembling

Benefits

The following are some of the benefits of Tabular Workflow for Forecasting:

• Supportslarge datasets that are up to 1TB in size and have up to 200 columns.
• Lets youimprove stability and lower training time by limiting the search space of architecture types or skipping architecture search.
• Lets youimprove training speed by manually selecting the hardware used for training and architecture search.
• Lets youreduce model size and improve latency by changing the ensemble size.
• Each component can be inspected in a powerful pipelines graph interface that lets you see the transformed data tables, evaluated model architectures and many more details.
• Each component gets extended flexibility and transparency, such as being able to customize parameters, hardware, view process status, logs and more.

Forecasting on Vertex AI Pipelines

Tabular Workflow for Forecasting is a managed instance of Vertex AI Pipelines.

Vertex AI Pipelines is a serverlessservice that runs Kubeflow pipelines. You can use pipelines to automateand monitor your machine learning and data preparation tasks. Each step in apipeline performs part of the pipeline's workflow. For example,a pipeline can include steps to split data, transform data types, and train a model. Since stepsare instances of pipeline components, steps have inputs, outputs, and acontainer image. Step inputs can be set from the pipeline's inputs or they candepend on the output of other steps within this pipeline. These dependenciesdefine the pipeline's workflow as a directed acyclic graph.

Overview of pipeline and components

The following diagram shows the modeling pipeline for Tabular Workflow for Forecasting:

The pipeline components are:

1. feature-transform-engine: Performs feature engineering. SeeFeature Transform Engine for details.

2. training-configurator-and-validator: Validates the training configuration and generates the training metadata.

   Input:

   • instance_schema: Instance schema in OpenAPI specification, which describes the data types of the inference data.
   • dataset_stats: Statistics that describe the raw dataset. For example,dataset_stats gives the number of rows in the dataset.
   • training_schema: Training data schema in OpenAPI specification, whichdescribes the data types of the training data.

3. split-materialized-data: Splits the materialized data into a training set, an evaluation set, and a test set.

   Input:

   • materialized_data: Materialized data.

   Output:

   • materialized_train_split: Materialized training split.
   • materialized_eval_split: Materialized evaluation split.
   • materialized_test_split: Materialized test set.

4. calculate-training-parameters-2: Calculates the expected runtime durationforautoml-forecasting-stage-1-tuner.

5. get-hyperparameter-tuning-results -Optional: If you configure thepipeline to skip the architecture search, load the hyperparameter tuningresults from a previous pipeline run.

6. Perform model architecture search and tune hyperparameters (automl-forecasting-stage-1-tuner) or use the hyperparameter tuning resultsfrom a previous pipeline run (automl-forecasting-stage-2-tuner).

   • An architecture is defined by a set of hyperparameters.
   • Hyperparameters include the model type and the model parameters.
   • Model types considered are neural networks and boosted trees.
   • A model is trained for each architecture considered.

   Input:

   • materialized_train_split: Materialized training split.
   • materialized_eval_split: Materialized evaluation split.
   • artifact - Hyperparameter tuning results from a previous pipeline run.This artifact is an input only if you configure the pipeline to skip thearchitecture search.

   Output:

   • tuning_result_output: Tuning output.

7. get-prediction-image-uri-2: Produces the correct inference image URI based on themodel type.

8. automl-forecasting-ensemble-2: Ensembles the best architectures to produce a final model.

   Input:

   • tuning_result_output: Tuning output.

   Output:

   • unmanaged_container_model: Output model.

9. model-upload-2 - Uploads the model.

   Input:

   • unmanaged_container_model: Output model.

   Output:

   • model: Vertex AI model.

10. should_run_model_evaluation -Optional: Use the test set to calculate evaluation metrics.

What's next

• Train a model using Tabular Workflow for Forecasting.