PENDIGITS

Pendigit dissimilarity matrix, 10992 x 10992

    A = PENDIGITS(SET,OBJECTS,REPSET,DATATYPE)

Input
SET	Integer indicating pendigit data set
1	pendigits-orig (pendigits-orig_norm9-VectorCostFunction2)
2	pendigits45 (anglepen-AngleCostFunction45)
3	pendigits60 (anglepen-AngleCostFunction60)
4	pendigits69 (anglepen-AngleCostFunction69)
5	pendigits90 (anglepen-AngleCostFunction90)
6	pendigits-mean-angle (normalised mean of sets (2-5))
7	pendigits-mean-all (normalised mean of set (1-5))
OBJECTS	Either a string: 'train', 'test' or 'all', or  a vector of indices between 1 and 10992
REPSET	Either a string: 'train', 'test' or 'all', or  a vector of indices between 1 and 10992
DATATYPE	string with desired datatype, 'dset' or 'dfile'  for 'dataset' or 'datafile'.

Output
A	Resulting dataset or datafile

Description

This routines loads dissimilarities of one of the five Pendigits data set  as computed by Bunke and Spillman. The original data has been collected  by Alimoglu & Alpaydin The digits are written by 44 different writers.  The data was originally split in a fixed training set of 7494 digits and a  fixed testset of 3498 digits. By setting OBJECTS and REPSET to 'train' or  'test' the corresponding subset of the dissimilarity matrix is taken.  Alternatively a subset may be defined by indices.

The sets 6 and 7 are only available as datasets for training and testing.  The full dissimilarity matrix ('all'), is not available.

Link(s)

http://www.iam.unibe.ch/fki/databases/string-edit-distance-matrices/  http://archive.ics.uci.edu/ml/machine-learning-databases/pendigits/  http://www.cmpe.boun.edu.tr/~alimoglu/tainn96.ps.gz

Reference(s)

D. Spillman, M. Neuhaus, H. Bunke, E. Pekalska, and R.P.W. Duin, Transforming Strings to Vector Spaces Using Prototype Selection, SSSPR2006, LNCS 4109, 2006, 287-296. X. Jiang and K. Abegglen and H. Bunke and J. Csirik, Dynamic computation of generalised median strings, Pattern Analysis and Applications, Vol. 6, pp. 185-193, 2003 F. Alimoglu, E. Alpaydin, Methods of Combining Multiple Classifiers Based on Different  Representations for Pen-based Handwriting Recognition, TAINN 96, June 1996, Istanbul, Turkey.

See also

datasets, datafiles,