Database Unbounded Array Framework Time Series Functions - Database Unbounded Array Framework Time Series Functions - Teradata Vantage

Using Unbounded Array Framework

Unbounded Array Framework (UAF) is the Teradata framework for building end-to-end time series forecasting pipelines. It also provides functions for digital signal processing and 4D spatial analytics. The series can reside in any Teradata supported or Teradata accessible table or in an analytic result table (ART).

Why Would I Use this Content?

Use this guide to learn about the functions that enable you to access your stationary datasets, then invoke and plot the results with autocorrelation, partial auto-correlation, and power spectrum functions. You can also access non-stationary datasets for trend removal, non-stationary variance (heteroscedastic) correction, and non-stationary properties of covariance, including removal of periodicities from the dataset.

How Do I Use this Content?

You use this guide as a reference for preparing data, creating forecasting models and determining goodness-of-fit of time series.

How Do I Get Started?

Learn about the series, matrixes, analytic result tables (ART) and generated series that are used with UAF functions, then learn how to prepare and use them to create forecasting models. See Unbounded Array Framework Specifications.

Unbounded Array Framework Overview

UAF functions process one-dimensional series and two-dimensional arrays. A one-dimensional series is a series indexed by time or space. A two-dimensional array is a matrix indexed by time, space or time and space. The indexing is discrete, meaning that the series or matrixes represent points occurring at equally-spaced intervals, with no intervening empty entries. These discrete indexes can be temporal, meaning they are indexed by date or timestamp, or number- or sequence-based, meaning they are indexed by integers or float-point numbers.

UAF supports series instance and matrix instance enabling any number of series instances or matrix instances to be stored in any supported Teradata table type or Teradata accessible table type. The instances are stored in the database as opposed to being stored in memory.

The elements can be the following types of numbers:

• Real numbers
• Complex numbers (rectangular form)
• Amplitude-phase pairs (polar form)
• Vectors of complex numbers
• Vectors of amplitude-phase pairs
• Mix of Teradata supported data types

The functions enable you to access your stationary datasets, then invoke and plot the results with auto-correlation, partial auto-correlation, and power spectrum functions. You can also access non-stationary datasets for trend removal, non-stationary variance (heteroscedastic) correction, and non-stationary properties of covariance, including removal of periodicities from the dataset.

The forecasting functions provide statistical test functions that enable you to qualify your models, and perform quality control tests against published models. The generated result sets are returned in natural order. Individual series are returned intact and in-order. Individual matrixes are returned intact in either row-major or column-major order. Results can be a table or an analytic result table (ART).

An ART can have multiple layers. Each layer possessing its own dedicated row composition for the series or matrix. For example, the TD_LINEAR_REGR (linear regression) function outputs a three-layer ART. The primary layer contains the computed coefficients. The secondary layer contains the goodness-of-fit metrics. The tertiary layer contains the residual series.

• Unbounded Array Framework Specifications
• General Utility Functions
• Data Preparation Functions
• Diagnostic Statistical Test Functions
• Model Preparation and Parameter Estimation Functions
• Series Forecasting Functions
• Temporal and Spatial Functions
• Mathematic Operators and Functions
• UAF Analytic Functions Examples