PRTools Contents

PRTools User Guide

prdataset

PRDATASET

Dataset class constructor

     A = PRDATASET(DATA,LABELS)

Input
 DATA size [M,K] a set of M datavectors (objects) of length K.  a cell array of datasets will be concatenated.
 LABELS size [M,N] array with labels for the M datavectors.  They should be either integers or character strings.  Choose single characters for the fastest implementation.  Numeric labels with value NaN or character labels  with value CHAR(0) are interpreted as missing labels.

Output
 A Dataset

Description

This command is the class constructor for datasets. In addition to the object labels  various other types of information can be stored in the fields of A.  These fields are

 DATA size [M,K] array (doubles) with M K-dimensional feature vectors (objects)
 FEATLAB size [K,F] array with labels for the K features
 FEATDOM size [K] cell array with domain description for the K features
 TARGETS size [M,C] dataset with soft labels or targets
 PRIOR size [C,1] prior probabilities for each of the C classes
  • PRIOR = 0: all classes have equal probability 1/C
  • PRIOR = []: all datavectors are equally probable
 COST size [C,C+1] Classification cost matrix. COST(I,J) are the costs  of classifying an object from class I as class J.  Column C+1 generates an alternative reject class and  may be omitted, yielding a size of [C,C].  An empty cost matrix, COST = [] (default) is interpreted  as COST = ONES(C) - EYE(C) (identical costs of  misclassification).
 LABLIST size [C,N] class labels corresponding to the unique labels found  in LABELS and thereby to the classes in the dataset.  The order of the items in LABLIST corresponds to the  apriori probablities stored in PRIOR. LABLIST should  only be given explicitely if PRIOR is given and if it  is not equal to 0 and not empty.
 LABTYPE String defining the label type,  'crisp' for defining classes by integers or strings  'soft' for defining memberships to classes. In this  case LABELS should be a MxC array with numbers  between 0 and 1.  'targets' for defining regression type target values.  Labels should be a MxN numeric array for  defining N targets per object.
 OBJSIZE number of objects, or vector with its shape. This is  useful if the set of objects can be interpreted as an  image (objects are pixels).
 FEATSIZE number of features, or vector with its shape. This is  useful if the set of features can be interpreted as an  image (features are pixels).
 IDENT [M,1] Cell array, identifier for objects.
 NAME String with dataset name
 USER User definable variable
 VERSION Date and PRTOOLS version at creation

The fields LABLIST, OBJSIZE, FEATSIZE, IDENT and VERSION are preset by PRTOOLS.  The other fields can be set by the user by the below SET commands.  All fields can be read by GET commands. By STRUCT(A) a dataset A can be  converted to a structure. By DOUBLE(A) or +A the data can be retrieved.  HELP DATASETS lists more information.

See also

datasets, mappings,

PRTools Contents

PRTools User Guide

This file has been automatically generated. If badly readable, use the help-command in Matlab.