Forecasting with ARIMA+

Stay organized with collections Save and categorize content based on your preferences.
Colab logo Run in Colab GitHub logo View on GitHub Vertex AI logo Open in Vertex AI Workbench

There are two options for time series forecasting models on Google Cloud:

Forecasting Model Comparison

The multivariate Vertex AI Forecasting model has the potential to model the data more effectively than the univariate BigQuery ML ARIMA_PLUS model because it can explicitly model covariates. For an inventory planning application, example covariates could be details about product advertisements, product attributes, holidays, and locations.

In general, it takes less time to train the BigQuery ML ARIMA_PLUS model than the Vertex AI Forecasting model. Training a BigQuery ML ARIMA_PLUS model is a good idea if you need to perform many quick iterations of model training or if you need an inexpensive baseline to measure other models against.

Tutorial Overview

This tutorial compares the performance of the two models on a synthetic dataset of product sales. By training a forecasting model on this kind of dataset, a store planner could determine how much inventory they need to order for each of their products and stores.

The main tutorial steps are as follows:

  1. Train the BigQuery ML ARIMA_PLUS model using an instance of Vertex AI Pipelines from Google Cloud Pipeline Components (GCPC).
  2. View the BigQuery ML ARIMA_PLUS model evaluation.
  3. Make a batch prediction with the BigQuery ML ARIMA_PLUS model.
  4. Create a Vertex AI Dataset resource.
  5. Train the Vertex AI Forecasting model.
  6. View the model evaluation.
  7. Make a batch prediction with the model.