TSPAC Homepage
Prof. Stanley L. Sclove (312) 9962681
IDS Dept. (MC 294) (312) 9962676
UIC
601 S. Morgan St.
Chicago, IL 606077124
slsclove@uic.edu
www.uic.edu/~slsclove
TSPAC (Time Series Segmentation Package)
is a package of computer programs for timeseries segmentation.
The current TSPAC programs use onestep ahead Bayesian classification.
This is a greedy algorithm which is suboptimal.
The programs will be further developed to include the Viterbi
algorithm.
Some further documentation on the algorithm
for TSPAC is available: ASCII text version
MS Word version.
TSPAC programs
The programs are classified according to
 the number of variables
 1 or p
 the mode
 automatic: a range of values for K, the number of states, is input; not automatic: a single value of
K is input
 the transition matrix
 sparse (transition allowed only to adjacent states) or not
Links to source code of individual programs

TSSG1CMA
 1 variable, common
variance, automatic mode

TSSG1CAS
 1 variable, common
variance, automatic mode, sparse transition matrix

TSSG1DT
 1 variable, different variances

TSSG1DTA
 1 variable, different variances,
automatic mode

TSSGPDTA
 P variables, different covariance matrices,
automatic mode
TSPAC is part of a suite of programs called
ISOPAC.
The other components of the suite are

CLUSPAC for
cluster analysis

IMPAC for digitalimage segmentation.
Created 1999: Dec 13
Latest update: 2003: Jul 19