realhrr: Implementation of Holographic Reduced Representations

Description Usage Arguments Details Value Author(s) References See Also Examples

Description

Specific functions for creating and working with Holographic Reduced Representations. Supplies functions for all standard algebraic VSA operations on vectors: multiplication, addition, superposition, scaling, powers (including inverses), approximate inverse, and similarity. These work in functional and operator form (e.g., add(x, y) and x * y both call the add method for class realhrr.

Usage

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
## S3 method for class 'realhrr'
newVec(what = c("rand", "I", "1", "0", "NA"), len = options("vsalen")[[1]],
        elts = NULL, cnorm = getOption("vsacnorm", TRUE), opnorm = getOption("vsaopnorm", FALSE),
        vsatype = getOption("vsatype"))
## S3 method for class 'realhrr'
vsaprod(e1, e2, method = c("fft", "outer"))
## S3 method for class 'realhrr'
appinv(e1)


## S3 method for class 'realhrr'
add(e1, ...)
## S3 method for class 'realhrr'
vnorm(e1)
## S3 method for class 'realhrr'
dot(e1, e2)
## S3 method for class 'realhrr'
equiv(e1, e2, tol)
## S3 method for class 'realhrr'
cosine(e1, e2, mag1, mag2)
## S3 method for class 'realhrr'
mag(e1, actual)
## S3 method for class 'realhrr'
vsapower(e1, e2)
## S3 method for class 'realhrr'
vsascale(e1, e2)
addnorm(...)

Arguments

what

the type of vector wanted: one of rand, I, 1, 0, NA:

  • rand is a vector filled with random samples from a Gaussian distribution with mean 0 and standard deviation 1/len

  • I and 1 mean the same thing and return the identity vector (i.e., the vector satisfying x * I = x for all non-zero x)

  • 0 is the zero vector, i.e., the vector satisfying x * 0 = 0 and x + 0 = x for all vectors x

  • NA is a vector filled with NA values.

what is ignored if elts is supplied.

len

the number of elements in the vector

elts

the raw elements for the vector

cnorm

a logical value (TRUE or FALSE) - controls whether the new vectors is created to have a magnitude of 1

opnorm

a logical value (TRUE or FALSE) - controls whether the result of a vsa operation is normalized to have a magnitude of 1

vsatype

ignored for newVec.realhrr (as with all methods of newVec)

e1

a vsa vector – a subclass of vsa

e2

a vsa vector for vsaprod, dot, equiv, cosine, and add, or a scalar (a numeric vector of length 1) for vsapower and vsascale

mag1,mag2

precomputed magnitudes for e1 and e2

actual

work with data values from this object (in this mode of invocation e1 is a template holding the class)

method

computational method for convolution (vsaprod)

tol

numerical tolerance for equiv

...

addtional vsa vectors for add

Details

These functions provide implementations of VSA operations for real-valued Holographic Reduced Representations (in the "spatial" domain). All of these functions take as arguments one or more realhrr VSA vectors, and possibily a scalar. The result is a realhrr VSA vector, or a scalar with class simval. To catch some errors due to misunderstanding of R's precedence, a simval scalar cannot be supplied as the argument of one of these functions (to do, remove the simval class using the function scalar).

Value

vsaprod, appinv, add, vnorm, vsapower, vsascale return a vsa object of the same class as their argument.

mag, dot and cosine return a scalar value with class simval

Author(s)

Tony Plate tplate@acm.org

References

http://www.d-reps.org

See Also

vsa, vsa.mem, for operator precedence: Syntax

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
library(vsa)
a <- newVec()
a
b <- newVec()
b
dot(a, b)
a 
elts(a)[1:10]
x <- a * b
(!a)
(!a * x) 
((!a) * x) 
(appinv(a) * x) 
(x / a) 
c <- newVec()
d <- newVec()
e <- newVec()
f <- newVec()
x <- vnorm(a * b + c * d + e * f)
mem <- vsamem(a, b, c, d, e, f)
cosmem(mem, a)
dotmem(mem, x / b)
cosmem(mem, x / b)
cosmem(mem, x / c)
cosmem(mem, x / f)
bestmatch(mem, x / b, cos=TRUE)
bestmatch(mem, x / b, cos=FALSE)

vsa documentation built on May 21, 2017, 12:47 a.m.

Search within the vsa package
Search all R packages, documentation and source code