PRED

Pairwise Relative Distance for Class-Vector or Class-Sample datasets

Pairwise Relative Distance for Class-Vector or Class-Sample datasets

S = pred(DATA) calculates pred for Class-Vector datasets.

DATA can be in one of the following formats:

- 2-d cell matrix with scalar or row vector values

- 2-d or 3-d double matrix. For 3-d matrices, values along the 3rd

dimension are used as vector values for each Class and Vector

combination.

There must be at least 2 rows and 2 columns. Each column is

considered as a Class and each row is considered as a Vector.

S is a column vector of length n*(n-1)/2, where n = size(DATA, 1),

corresponding to all Vector pairs. Each value of S is averaged over all

m*(m-1)/2 pairs of Classes, where m = size(DATA, 2). NaNs are ignored

while calculating the mean.

S = pred(DATA, CLASS_LABELS) calculates pred for Class-Sample datasets.

DATA can be in one of the following formats:

- 1-d column cell vector with scalar or row vector values

- 1-D column or 2-D double matrix. For 2-d matrices, values along

the 2nd dimension are used as vector values for each Sample.

There must be at least 4 rows in data.

CLASS_LABELS specify the Class for each row in DATA and should be

either numeric or a cell of strings. There must be at least 2 Samples

per Class.

S is a row vector of length m(m-1)/2 corresponding to all Class pairs,

where m = number of Classes. For each pair of Classes i-j, S is

averaged over all n_i*(n_i-1)*n_j*(n_j-1)/2 pairs of Samples (Exhaustive

case) or all n*(n-1)/2 pairs of Samples (Fast case), where n_i, n_j =

number of Samples for Class i and j, respectively; n = max(n_i, n_j).

NaNs are ignored while calculating the mean.

S = pred(DATA, {}) calculates pred for Class-Vector datasets but

generates class_labels automatically as 1, 2, ... depending on the

column number. DATA can be specified in the same way as for

Class-Vector case. Each column is considered as a class and each row as

a Sample.

[S, LABELS] = pred(...) also returns LABELS, which is a cell matrix

specifying the class or vector pair for each value in S. length(LABELS)

is equal to length(D).

[...] = pred(..., 'PARAM1', VAL1, 'PARAM2', VAL2) specifies additional

parameters and their values. Valid parameters are the following:

Parameter Value

'distance' Specifies the distance function to be used for

calculation of distances D1 and D2 in the pred

function. 'squaredeuclidean' is used by default.

Other options are 'euclidean', 'seuclidean',

'cityblock', 'chebychev' which have the same meaning

as specified in the pdist function

Only for Class-Sample case:

'type' Either 'exhaustive' (default) or 'fast'. 'exhaustive'

mode considers all possible Sample pairs per Class

for calculation and provides a more accurate

description of data. 'fast' mode only considers a

subset of the Sample pairs and should only be used if

number of Samples is large.