From: Dave Watson <watson@maths.uwa.edu.au>
Newsgroups: comp.os.linux.announce,sci.math.num-analysis,sci.misc,sci.geo.geology,comp.theory.dynamic-sys
Subject: Time-series prediction
Date: 1 Feb 1994 11:54:05 +0200
Approved: linux-announce@tc.cornell.edu (Lars Wirzenius)
Message-ID: <2il8rt$24g@klaava.Helsinki.FI>

                Time-series Prediction

A package, k_oz,  which will generate predictions from a set of
time-series data is freely available by ftp from -

madvax.uwa.edu.au    /pub/dddd/*

k_oz will read a set of time-series data, embed it in a specified
dimension with a specified lag, generate a simplicial approximation
to the embedding manifold, and perform interpolation within the 
simplicial complex to obtain the predictions.  Interpolation is by
the natural neighbor method or any of six radial basis functions
applied to a natural neighbor subset.

The data may be divided into training and testing subsets and the 
coefficient of linear correlation of the predictions against the 
training set may be obtained.  This may indicate the existence of 
a dynamic deterministic source mechanism for the data.  Three
techniques for estimating embedding dimension are provided.

If the embedding is in 2D or 3D, the data, or a sequential trace 
of the data, or the triangular facets of the simplicial 
approximation, may be displayed as a rotatable stereogram with
zoom, pan, or slow-motion construction.

The program will compile and run on a machine with OPENWINDOWS
version 3, and with the XView version 3.0 libraries in directory 
$OPENWINHOME.  It was built on a Sun SPARC and I run it with
Linux (SLS 1.02) on my 486DX33 at home.

The example data set is 101 clay varve thicknesses from the
geological literature, and cascaded predictions based on the 
first half of the data have a positive but gradually declining 
correlation coefficient up to eighteen steps when embedded in 
5D with a lag of eight.

If you try it out, we would appreciate hearing about your
impressions, comments, or criticism.
--
Dave Watson                          Internet: watson@maths.uwa.edu.au
Department of Mathematics                         Tel: (61 9) 380 3359
The University of Western Australia               FAX: (61 9) 380 1028
Nedlands, WA 6009  Australia.          Real data are full of surprises

--
Mail submissions for comp.os.linux.announce to: linux-announce@tc.cornell.edu
PLEASE remember Keywords: and a short description of the software.
