Dist2Mat conversion

Description

Efficiently converts a "dist" object into a symmetric distance "matrix".

Usage

Dist2Mat(dist)

Arguments

Details

Converts a "dist" object, typically created using the stats::dist function, into a symmetric matrix form. This implementation is optimised for speed and performs significantly faster than base::as.matrix or proxy::as.matrix when applied to "dist" objects.

Row names are retained. If it is null, as.character(1:nObs) will be used as the row and column names of the resulting matrix instead.

Value

A distance "matrix".

Author(s)

Minh Long Nguyen edelweiss611428@gmail.com

Examples

library("microbenchmark")
x = matrix(rnorm(200), nrow = 50)
dx = dist(x)
#Dist2Mat conversion
microbenchmark(base::as.matrix(dx),
               proxy::as.matrix(dx),
               Dist2Mat(dx))
#Check if equal
v1 = as.vector(base::as.matrix(dx))
v2 = as.vector(Dist2Mat(dx))
all.equal(v1, v2)

Expanding a distance matrix given new data

Description

Efficiently appends new "rows" to an existing "dist" object without explicitly recomputing a full pairwise distance matrix.

Usage

expandDist(distA, A, B, method = "euclidean", diag = FALSE, upper = FALSE, p = 2)

Arguments

Details

Expands an existing distance matrix of class "dist" for matrix A, given new data B, without explicitly computing the distance matrix of rbind(A,B). This supports multiple commonly used distance measures and is optimised for speed.

Row names are retained. If either rownames(A) or rownames(B) is null, as.character(1:(nrow(A)+nrow(B))) will be used as row names instead.

Value

A distance matrix of class "dist" for rbind(A,B).

Author(s)

Minh Long Nguyen edelweiss611428@gmail.com

Examples

A = matrix(rnorm(100), nrow = 20)
B = matrix(rnorm(250), nrow = 50)
AB = rbind(A,B)
distA = fastDist(A)
v1 = as.vector(expandDist(distA, A, B))
v2 = as.vector(fastDist(AB))
all.equal(v1, v2)

"dist" object computation

Description

Efficiently computes a "dist" object from a numeric matrix using various distance metrics.

Usage

fastDist(X, method = "euclidean", diag = FALSE, upper = FALSE, p = 2L)

Arguments

Details

Calculates pairwise distances between rows of a numeric matrix and returns the result as a compact "dist" object, which stores the lower-triangular entries of a complete distance matrix. Supports multiple distance measures, including "euclidean", "manhattan", "maximum", "minkowski", "cosine", and "canberra". This implementation is optimised for speed, especially on large matrices.

Row names are retained. If it is null, as.character(1:nrow(X)) will be used as row names instead.

Value

A distance matrix of class "dist".

Author(s)

Minh Long Nguyen edelweiss611428@gmail.com

Examples

library("microbenchmark")
x = matrix(rnorm(200), nrow = 50)
microbenchmark(stats::dist(x, "minkowski", p = 5),
               fastDist(x, "minkowski", p = 5))
v1 = as.vector(stats::dist(x, "minkowski", p = 5))
v2 = as.vector(fastDist(x, "minkowski", p = 5))
all.equal(v1, v2)

Computing pairwise distances between rows of two matrices

Description

Efficiently computes pairwise distances between the rows of two numeric matrices using various distance metrics.

Usage

fastDistAB(A, B, method = "euclidean", p = 2L)

Arguments

Details

This function computes the full pairwise distance matrix between the rows of matrices A and B, without forming a concatenated matrix or performing unnecessary intermediate conversions. It supports multiple commonly used distance measures and is optimised for speed.

Row names in A and B are retained. If either rownames(A) or rownames(B) is null, as.character(1:nrow(A)) and as.character(1:nrow(B)) will be used as row and column names of the resulting matrix instead.

Value

A numeric matrix of dimensions nrow(A) by nrow(B), where each entry represents the distance between a row in A and a row in B.

Author(s)

Minh Long Nguyen edelweiss611428@gmail.com

Examples

library("microbenchmark")
X = matrix(rnorm(200), nrow = 50)
A = X[1:25,]
B = X[26:50,]
microbenchmark(proxy::dist(A,B, "minkowski", p = 5),
               fastDistAB(A,B, "minkowski", p = 5L))
#Check if equal
v1 = as.vector(proxy::dist(A,B, "minkowski", p = 5))
v2 = as.vector(fastDistAB(A,B, "minkowski", p = 5L))
all.equal(v1, v2)

2D-indexing to 1D-indexing

Description

Efficiently computes 1D-indexing from 2D-indexing

Usage

get1dFrom2d(i,j, N)

Arguments

Details

Converts 2D indexing (a row-column pair) into 1D indexing (as used in R's "dist" objects), given the number of observations N.

Currently, name-based indexing is not supported."

Value

An integer specifying the 1d index

Author(s)

Minh Long Nguyen edelweiss611428@gmail.com

Examples

N = 5
for(i in 1:4){
  for(j in (i+1):5){
    print(get1dFrom2d(i,j,N))
  }
}

1D-indexing to 2D-indexing

Description

Efficiently computes 2D-indexing from 1D-indexing

Usage

get2dFrom1d(idx1d, N)

Arguments

Details

Converts 1D indexing (as used in R's "dist" objects) into 2D indexing (row-column pairs) for a distance matrix of size N \times N.

Currently, name-based indexing is not supported."

Value

An integer matrix storing the corresponding 2D indexes.

Author(s)

Minh Long Nguyen edelweiss611428@gmail.com

Examples

get2dFrom1d(1:10, 5)

Subsetting a "dist" object by columns

Description

Efficiently extracts a column-wise subset of a "dist" object, returning the corresponding submatrix of pairwise distances. # nolint

Usage

subCols(dist, idx)

Arguments

Details

This function extracts specified columns from a "dist" object without explicit conversion to a dense distance "matrix", resulting in better performance and reduced memory overhead. Particularly useful when only a subset of distances is needed for downstream tasks.

Row names are retained. If it is null, as.character(1:nObs) and as.character(idx) will be used as row and column names of the resulting matrix instead.

Value

A numeric "matrix" containing the pairwise distances between all rows and the specified columns.

Author(s)

Minh Long Nguyen edelweiss611428@gmail.com

Examples

library("microbenchmark")
x = matrix(rnorm(200), nrow = 50)
dx = dist(x)
#Randomly subsetting a 50x10 matrix
idx = sample(1:50, 10)
microbenchmark(base::as.matrix(dx)[1:50,idx],
               proxy::as.matrix(dx)[1:50,idx],
               subCols(dx, idx))
#Check if equal
v1 = as.vector(base::as.matrix(dx)[1:50,idx])
v2 = as.vector(subCols(dx, idx))
all.equal(v1, v2)

Dist2Dist subsetting

Description

Efficiently extracts a subset of observations from a "dist" object and returns a new "dist" object representing only the selected distances.

Usage

subDist2Dist(dist, idx, diag = FALSE, upper = FALSE)

Arguments

Details

This function subsets a "dist" object directly without explicit conversion to a dense distance "matrix". It extracts only the relevant distances corresponding to the selected indices, improving both performance and memory efficiency. The result is returned as a subsetted "dist" object, preserving compatibility with downstream functions that accept this class.

Row names are retained. If it is null, as.character(idx) will be used as row names instead.

Value

A numeric "matrix" storing pairwise distances between the selected observations.

Author(s)

Minh Long Nguyen edelweiss611428@gmail.com

Examples

library("microbenchmark")
x = matrix(rnorm(200), nrow = 50)
dx = dist(x)
#Subsetting the first 10 units
microbenchmark(as.dist(base::as.matrix(dx)[1:10,1:10]),
               as.dist(proxy::as.matrix(dx)[1:10,1:10]),
               subDist2Dist(dx, 1:10))
#Check if equal
v1 = as.vector(as.dist(base::as.matrix(dx)[1:10,1:10]))
v2 = as.vector(subDist2Dist(dx, 1:10))
all.equal(v1, v2)

Dist2Mat subsetting

Description

Efficiently extracts a 2d submatrix of pairwise distances from a "dist" object.

Usage

subDist2Mat(dist, idx1, idx2)

Arguments

Details

This function efficiently subsets a "dist" object by row and column indices, returning the corresponding rectangular section as a numeric matrix. It avoids explicit conversion from the "dist" object to a dense "matrix", improving memory efficiency and computational speed, especially with large datasets.

Row names are retained. If it is null, as.character(idx1) and as.character(idx2) will be used as row and column names of the resulting matrix instead.

Value

A numeric matrix storing pairwise distances between observations column-indexed by idx1 and row-indexed by idx2.

Author(s)

Minh Long Nguyen edelweiss611428@gmail.com

Examples

library("microbenchmark")
x = matrix(rnorm(200), nrow = 50)
dx = dist(x)
#Randomly subsetting a 10x10 matrix
idx1 = sample(1:50, 10)
idx2 = sample(1:50, 10)
microbenchmark(base::as.matrix(dx)[idx1,idx2],
               proxy::as.matrix(dx)[idx1,idx2],
               subDist2Mat(dx, idx1, idx2))
#Check if equal
v1 = as.vector(base::as.matrix(dx)[idx1,idx2])
v2 = as.vector(subDist2Mat(dx, idx1, idx2))
all.equal(v1, v2)