Overview

Store Sales Forecast User Interface Tutorial

This tutorial series walks you through the process of setting up a forecasting catalog using the User Interface. You'll learn how to:

• Create a catalog for a time series forecasting use case.
• Register tables including a Time Series table with timezone support and a Calendar table.
• Define entities and their relationships.
• Formulate a forecasting use case with a regression target.
• Create observation tables using forecast automation.
• Use Ideation to generate features and models for forecasting.

Dataset Overview

In this tutorial, we work with the Store Sales Forecast dataset, derived from the M5 Forecasting Accuracy Kaggle competition. The dataset is aggregated at the store level with sales_amount (revenue = sales × sell_price) as the target.

The dataset is composed of three tables:

• SALES: Daily total sales amount (revenue) per store, with an IANA timezone column.
• CALENDAR: Per-state daily calendar with event names and a unified SNAP eligibility flag.
• STORE_STATE: Mapping from store to US state.

Entity Hierarchy

store_id  ──→  state_id

state_id links stores to the per-state calendar (events and SNAP flags).

Tutorial Structure

This tutorial follows a structured, end-to-end workflow:

1. Create Catalog

Define the Data Model

2. Register Tables

3. Register Entities

Skipped Steps

This tutorial skips setting default cleaning operations, updating descriptions, and tagging semantics to focus on the forecasting workflow. For guidance on these recommended steps, see the Credit Default UI tutorials: Set Default Cleaning Operations and Update Descriptions and Tag Semantics.

Formulate your Use Case

4. Formulate Use Case

5. Create Observation Tables

Ideate Features and Models

6. Ideate Features and Models

Predict and Evaluate

7. Predict and Evaluate