datasets

DATASETS

Info on the dataset class construction for PRTools

This is not a command, just an information file.

Datasets in PRTools are in the MATLAB language defined as objects of the class PRDATASET. Below, the words 'object' and 'class' are used in the pattern recognition sense.

A dataset is a set consisting of M objects, each described by K features. In PRTools, such a dataset is represented by a M x K matrix: M rows, each containing an object vector of K elements. Usually, a dataset is labeled. An example of a definition is

    DATA = [RAND(3,2) ; RAND(3,2)+0.5];
    LABS = ['A';'A';'A';'B';'B';'B'];
    A = PRDATASET(DATA,LABS)

which defines a [6 x 2] dataset with 2 classes.

The [6 x 2] data matrix (6 objects given by 2 features) is accompanied by labels, assigning each of the objects to one of the two classes A and B. Class labels can be numbers or strings and should always be given as rows in the label list. A lable may also have the value NaN or may be an empty string, indicating an ulabeled object. If the label list is not given, all objects are marked as unlabeled.

Various other types of information can be stored in a dataset. The most simple way to get an overview is by typing

STRUCT(A)

which for the above example displays the following

           DATA: [6x2 double]
        LABLIST: [2x1 double]
           NLAB: [6x1 double]
        LABTYPE: 'crisp'
        TARGETS: []
        FEATLAB: [2x1 double]
        FEATDOM: {1x2 cell}
          PRIOR: []
           COST: []
        OBJSIZE: 6
       FEATSIZE: 2
          IDENT: {6x1 cell}
        VERSION: {1x2 cell}
           NAME: []
           USER: []

These fields have the following meaning

DATA an array containing the objects (the rows) represented by features (the columns). In the software and help-files, the number of objects is usually denoted by M and the number of features is denoted by K. So, DATA has the size of [M,K]. This is also defined as the size of the entire dataset.
LABLIST The names of the classes, stored row-wise. These class names should be integers, strings or cells of strings. Mixtures of these are not supported. LABLIST has as many rows as there are classes. This number is usually denoted by C. LABLIST is constructed from the set of LABELS given in the DATASET command by determining the unique names while ordering them alphabetically.
NLAB an [M x 1] vector of integers between 1 and C, defining for each of the M objects its class. They are indexing LABLIST.
LABTYPE 'CRISP', 'SOFT' or 'TARGETS' are the three possible label types. In case of 'CRISP' labels, a unique class, defined by NLAB, is assigned to each object, pointing to the class names given in LABLIST. For 'SOFT' labels, each object has a corresponding vector of C numbers between 0 and 1 indicating its membership (or confidence or posterior probability) of each of the C classes. These numbers are stored in the array TARGETS of the size M x C. They don't necessarily sum to one for individual row vectors. Labels of type 'TARGETS' are in fact no labels, but merely target vectors of length C. The values are again stored in TARGETS and are not restricted in value.
TARGETS [M,C] array storing the values of the soft labels or targets.
FEATLAB A label list (like LABLIST) of K rows storing the names of the features.
FEATDOM A cell array describing for each feature its domain.
PRIOR Vector of length C storing the class prior probabilities. They should sum to one. If PRIOR is empty ([]) it is assumed that the class prior probabilities correspond to the class frequencies.
COST 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).
OBJSIZE The number of objects, M. In case the objects are related to a n-dimensional structure, OBJSIZE is a vector of length n, storing the size of this structure. For instance, if the objects are pixels in a [20 x 16] image, then OBJSIZE = [20,16] and M = 320.
FEATSIZE The number of features, K. In case the features are related to an n-dimensional structure, FEATSIZE is a vector of length n, storing the size of this structure. For instance, if the features are pixels in a [20 x 16] image, then FEATSIZE = [20,16] and K = 320.
IDENT A cell array of M elements storing indicators of the M objects. They are initialized by integers 1:M.
VERSION Some information related to the version of PRTools used for defining the dataset.
NAME A character string naming the dataset, possibly used to annotate related graphics.
USER Free field for the user, not used by PRTools.

The fields can be set by commands like SETDATA, SETFEATLAB, SETLABELS, see below for a complete list. Note that there is no field LABELS in the DATASET definition. Labels are converted to NLAB and LABLIST. The command SETLABELS however exists and takes care of the conversion.

The data and information stored in a dataset can be retrieved as follows

By DOUBLE(A) and by +A, the content of A.DATA is returned.
[N,LABLIST] = CLASSSIZES(A); It returns the numbers of objects per class and the class names stored in LABLIST.
By DISPLAY(A), it writes the size of the dataset, the number of classes and the label type on the terminal screen.
By SIZE(A), it returns the size of A.DATA: numbers of objects and features.
By SCATTERD(A), it makes a scatter plot of a dataset.
By SHOW(A), it may be used to display images that are stored as features or as objects in a dataset.
By commands like: GETDATA, GETFEATLAB, etcetera, see below. With some exceptions they point to a single dataset field. E.g. GETSIZE(A) returns [M,K,C]. A aet of commands does not return data, but instead they return indices to objects that have specific identifiers, labels or class indices
FINDIDENT, FINDLABELS, FINDNLAB.

Many standard MATLAB operations and a number of general MATLAB commands have been overloaded for variables of the DATASET type.

Info on the dataset class construction for PRTools

See also