1.0 - 8.00 - Time Series, Path, and Attribution Analysis - Teradata Vantage

Teradata® Vantage Machine Learning Engine Analytic Function Reference

Teradata Vantage
Release Number
Release Date
May 2019
Content Type
Programming Reference
Publication ID
English (United States)
Function Description
ARIMA Functions Create and use ARIMA model.
Attribution Calculates attributions with a wide range of distribution models. Often used in web-page analysis.
Burst Bursts (splits) a time interval into a series of shorter "burst" intervals and allocates values from the time intervals into the new, shorter subintervals.
Change-Point Detection Functions Detect the change points in stochastic process or time series.
Convergent Cross-Mapping Functions Includes CCMPrepare function, which adds new partition column and partitions data to input to CCM function, which tests multiple causes and effects simultaneously, reporting effect size for each cause-effect pair.
DTW Computes dynamic time warping—similarity between two sequences that vary in time or speed.
Fast Fourier Transform Functions FFT uses Fast Fourier Transform (FFT) algorithm to compute discrete Fourier Transform (DFT) of each signal in one or more input table columns. Signal can be either real or complex, and can have one, two, or three dimensions.

IFFT uses inverse Fast Fourier Transform (IFFT) algorithm (also called Fourier synthesis algorithm) to reverse Fast Fourier Transform (FFT) performed by FFT function.

FrequentPaths Finds patterns that appear more than specified number of times in sequence database. Difference between sequential pattern mining and frequent pattern finding is that the former works on time sequences where order of items must be kept.
Interpolator Calculates missing values in time series, using either interpolation or aggregation. Interpolation estimates missing values between known values. Aggregation combines known values to produce aggregate value.
Path Analysis Functions Automate path analysis. Useful for clickstream analysis of web site traffic and other sequence/path analysis tasks, such as advertisement or referral attribution.
SAX Turns time series into symbolic strings, which are smaller and in which patterns are more obvious. Input and output formats allow SAX to supply data to Shapelet functions.
SeriesSplitter Splits partition into subpartitions (called splits) by creating additional column that contains split identifier. Optionally copies specified number of boundary rows to each split.
Sessionize Maps each click in clickstream to unique session identifier.
Shapelet Functions Detect distinguishing features among ordered sequences (time series) and use them to cluster or classify new data.
TimeSeriesOrders Determines orders and whether to incorporate drift term into time series model.
VARMAX Extends ARMA/ARIMA model to work with time series with multiple response variables (vector time series), as well as exogenous variables, or variables that are independent of other variables in system.
Wavelet Transform Functions Implement Mallat algorithm (iterate algorithm in Discrete Wavelet Transform field) and apply wavelet transform on multiple sequences simultaneously, or perform inverse of this process.