GENDDAT

Generate dissimilarity sets for training, testing and representation

   [DT,DS,I,J,R] = GENDDAT(D,M,K,L)
   [DT,DS,I,J,R] = GENDDAT(D,M,K,L,TYPE)

Input
D	NxN dissimilarity dataset
M	Vector of class sizes or frequencies, or a single number / frequency in(0,1), used for the trainset
K	Vector of class sizes or frequencies, or a single number / frequency in(0,1), used for the repset.(optional, default: K = M)
L	Vector of class sizes or frequencies, or a single number / frequency in(0,1), added tot the repset.(optional, default: L = [])
TYPE	String, 'in' or 'ex', default 'in'.

Output
DT	Training dissimilarity dataset, DT = D(I,R)
DS	Test dissimilarity dataset, DS = D(J,R)
I	Indices of the training objects
J	Indices of the test objects
R	Indices of the representation set

Description

This routine is similar to GENDAT for randomly splitting datasets, but it  randomly selects the representation set (the columns of DT and DS) as  well. The input matrix D (a PRTools dataset) should be square, listing  dissimilarities of a set of objects to themselves.

    M  determines the split of all objects in parts for training and testing.
       For M = [] all objects are used for DT.
    K  determines which part of training objects is used for representation.
       By default, K = [], all training objects are used. For K = 0 no
       objects from the trainset are used for the repset.
    L  determines which part of the test objects are used for representation.
       By default none of the test objects are used. In case TYPE is 'ex'
       the repset is kept outside testset.

For setting M, K and L the following rules apply (similar to GENDAT).

• a scalar smaller than 1 will generate a fraction of the set, using the  class priors of the dataset. Use 0.99999 for the entire set.
• a scalar equal or larger than 1 with generate that number of objects  using the class priors of the dataset.
• a vector with length equal to the number of classes will generate the  number (or fraction) of objects for the corresponding class, using the  class ranking as obtained by GETLABLIST(D).

Example(s)

 Let D be 100 x 100 dataset with two classes [40 60] and priors [0.4 0.6].
 1)  [DT,DS] = GENDDAT(D,0.6) 
 DTR is 60x60 and DTE is 40x60. 60% of the 40 objects of the first class 
 (24) and 60% of the 60 objects of the second class (36) are selected for
 DTR. The remaining objects are stores in DTE. The repset equals the
 trainset.

 2) [DT,DS] = GENDDAT(D,0.6,0.1) 
 DTR is 60x6 and DTE is 40x6. 60% of the 40 objects of the first class 
 (24) and 40% of the 60 objects of the second class (36) are selected 
 for training. From that, 10% of the 24 first class training objects are
 selected for the repset (rounded to 2). 10% of 36 second class objects,
 rounded to 4 objects for the second class are added.

 3) [DT,DS] = GENDDAT(D,0.6,0,0.1)
 DTR is 60x4 and DTE is 40x4. The 4 objects in the repset are taken from
 the testset.

 4) [DT,DS] = GENDDAT(D,0.6,0,0.1,'ex')
 DTR is 60x4 and DTE is 36x4. The 4 objects in the repset are exclcuded
 from the testset.

See also

datasets, getlablist, gendat, seldclass,