bpa: Basic Probability Assignment Objects

Description

These functions can be used to create, combine and print basic probability assignment objects for classificiation. A Basic Probabilty Assignment is a similar to a probabilty mass function, except that it has an additional mass for the concept for "ignorance".

Usage

bpa(n = 1, setlist = c(1:n, Inf), mlist = c(rep(0, n), 1))
"print"(x, verbose = FALSE, ...)

Arguments

The number of distinct values that need to be represented. Usually set to the number of classes in the data.

setlist

A subset of 1:n, indicating those elements that have positive mass. The special value 'Inf' is used to denote the whole set, which is ignorance in dempster-shafer terms.

mlist

The actual masses assigned to the elements in the setlist.

The bpa object to be printed.

verbose

If FALSE (default), simply prints out the basic probability assignment. If TRUE, prints a list of all member functions as well.

...

Additional arguments to print method. Not Used.

Value

get.N: Get the number of distinct values represented in the bpa. Usually set to the number of classes.
get.setlist: Get the sets represented in the bpa
get.full.m: Get the masses assigned to each of the elements.
get.focal.elements: Get only those elements that have a positive mass attached to them. Such elements are called the focal elements of the bpa.
get.m: Get the masses attached only to the focal elements, that is the non-zero elements of the mlist.
get.mass: Get the masses attached to certain specified elements of the bpa. The elements are specifed as a vector via the argument s.
assign.bpa: Can be used to re-assign mass to certain specified elements of the bpa. The argument s is the same as the setlist, and m is the same as mlist.
get.assigned.class: Returns a vector of all possible classes, in decreasing order of assigned probabilty. (That is, the first element is the most likely class, and the last element is the least likely class.)
get.assigned.ratios: Returns a vector of length N-1, each of whose elements is the ratio of the probability assigned to the successive classes. That is, the first element is the ratio of the probability assigned to the most likely class to that assigned to the next most likely class, and so on.
set.name: Can be passed a string to name the bpa object. (Used by the bpamat function; names are set to the row number.)
get.name: Returns the name of the bpa object. If not assigned, returns NULL.
set.info: Can be used to store as auxiliary information. (Used internally by the bpamat function to store the random seed, or the special character 'C' if the bpa object resulted from a combination.)
get.info: Returns whatever was stored as info. If empty, NULL is returned.

Details

It should be noted that these functions are fairly simplistic, since they were designed to be fast, and work with classification only. In particular, if you have set-valued elements, the combination function will likely give the wrong answer unless the sets are non-intersecting. For the same reason, belief functions have not been implemented either, since for atomic elements, bpa = belief function.

References

Gordon, J. and Shortliffe, E. H. (1984). The dempster-shafer theory of evidence. Rule-Based Expert Systems: The MYCIN Experiments of the Stanford Heuristic Programming Project, 3:832-838. Shafer, G. (1986). The combination of evidence. International Journal of Intelligent Systems, 1(3):155-179.

Examples

Run this code

##Empty bpa - All mass is attached to ignorance
b1 <- bpa(3)
b1

##Add a set to this bpa
b1$assign.bpa(s = c(1,2), m = c(0.3,0.4))
print(b1, verbose = TRUE)

##The same thing in a different way - classes can be named
##Note that the print method omits empty classes
b0 <- bpa(3, c('A','B','C', Inf), c(0.3, 0.4, 0, 0.3))
b0

##Another bpa
##Again, class '2' has been omitted
b2 <- bpa(3)
b2$assign.bpa(s = c(1,3), m = c(0.7,0.1))
b2

##Combine
b3 <- combine.bpa.ds(b1,b2)
b3
combine.bpa.bs(b1,b2)

##As a list, should be same answer as above
b4 <- combine.bpa.list.ds(list(b1,b2))
b4
combine.bpa.list.bs(list(b1,b2))

Run the code above in your browser using DataLab