Tapkee: Tapkee wrapper
In tapkee: Wrapper for 'tapkee' Dimension Reduction Library

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

View source: R/tapkee.r

R wrapper for the 'tapkee' dimension reduction library

1	Tapkee(data, method="pca", td=2, verbose=FALSE, add="", prefix="Dim", rm=TRUE)

`data`	R numerical matrix or data frame (will be converted into matrix)
`method`	'tapkee' method, run "system('tapkee -h')" for the list, default is "pca"
`td`	Number of dimensions to output, default is 2
`verbose`	If TRUE, 'tapkee' is verbose, defalut is FALSE
`add`	'tapkee' additional arguments as character string: see "system('tapkee -h')"
`prefix`	Variable name prefix in the resulted data frame, default is "Dim"
`rm`	Remove temp files (but temp folder will be removed anyway in the end of R session), default is TRUE

Interface (wrapper) for the 'tapkee', flexible and efficient C++ template library for dimension reduction. 'tapkee' is extremely fast comparing with other DR tools.

For methods used in 'tapkee', run 'vignette(tapkee_methods)'.

Users should install 'tapkee' independently from author Web site (https://github.com/lisitsyn/tapkee) or associated GitHub (https://github.com/lisitsyn/tapkee). Run 'package?tapkee' or help("tapkee-package") for details related with your operation system. If 'tapkee' is not installed, Tapkee() will fail gracefully and output the input data with warning.

Please note that "[warning] The neighborhood graph is not connected" message in most cases means that 'tapkee' run was unsuccessful. As a result, Tapkee() might return the matrix of NaN's. One of possible workarounds is to specify the higher number of neigbors ('-k' option, default is 10). See below for the example.

Note that the wrapper catches only one (main) type of 'tapkee' utility outputs. For other possible output types (see 'tapkee -h' for explanation), run 'tapkee' without wrapper.

Data frame with number of columns equal to number of dimensions given and "prefix" column names prefixes.

Alexey Shipunov

Sergey Lisitsyn and Christian Widmer and Fernando J. Iglesias Garcia. Tapkee: An Efficient Dimension Reduction Library. Journal of Machine Learning Research, 14: 2355-2359, 2013.

tapkee-package

## 'tapkee' vs. R base functions
system.time(Tapkee(iris[, -5], method="mds"))
system.time(cmdscale(dist(iris[, -5])))

## How to use 'add' option
plot(Tapkee(iris[, -5], "isomap", add="-k 47"), col=iris[, 5])

## 'tapkee' methods as of March 2019:
TM <- c(
"lle", # 1) locally_linear_embedding (lle),
"npe", # 2) neighborhood_preserving_embedding (npe),
"ltsa", # 3) local_tangent_space_alignment (ltsa),
"lltsa", # 4) linear_local_tangent_space_alignment (lltsa), 
"hlle", # 5) hessian_locally_linear_embedding (hlle),
"la", # 6) laplacian_eigenmaps (la),
"lpp", # 7) locality_preserving_projections (lpp),
"dm", # 8) diffusion_map (dm),
"isomap", # 9) isomap (isomap),
"l-isomap", # 10) landmark_isomap (l-isomap),
"mds", # 11) multidimensional_scaling (mds),
"l-mds", # 12) landmark_multidimensional_scaling (l-mds),
"spe", # 13) stochastic_proximity_embedding (spe),
"kpca", # 14) kernel_pca (kpca),
"pca", # 15) pca (pca),
"ra", # 16) random_projection (ra),
"fa", # 17) factor_analysis (fa),
"t-sne", # 18) t-stochastic_neighborhood_embedding (t-sne),
"ms") # 19) manifold_sculpting (ms)

## Iris example
oldpar <- par(mfrow=c(4, 5), mar=c(1, 1, 3, 1), xaxt="n", yaxt="n")
for (n in c(1:18)) {
plot(Tapkee(iris[, -5], method=TM[n], add="-k 50"),
 col=iris[, 5], pch=20, main=TM[n], xlab="", ylab="")
}
plot(iris[, 1:2], col=iris[, 5], pch=20, main="iris[, 1:2]", xlab="", ylab="")
par(oldpar)

## Generate typical 3D data
SR <- Gen.dr.data("swissroll")
SC <- Gen.dr.data("scurve")
HX <- Gen.dr.data("helix")
SS <- Gen.dr.data("ssphere")

## This will separate colors better
COL <- rainbow(1100)[1:1000]

## 3D plot (if no 'scatterplot3d' package, plots XY axes)
T3D <- function(dat, title, col=COL) {
 if (requireNamespace("scatterplot3d", quietly = TRUE)) {
 scatterplot3d::scatterplot3d(dat, color=col, main=title, pch=20, xlab="", ylab="", zlab="",
  axis=FALSE, tick.marks=FALSE, label.tick.marks=FALSE, mar=c(1, 1, 3, 1))
} else {
 plot(dat[, 1:2], col=col, main=paste(title, "(2D projection)"), pch=20)
 warning("Please install 'scatterplot3d' package to see the 3D plot")
}}

## Swiss Roll
oldpar <- par(mfrow=c(4, 5))
T3D(SR, title="Swiss Roll")
for (n in 1:18) plot(Tapkee(SR, method=TM[n]), col=COL, pch=20, main=TM[n],
 xlab="", ylab="", xaxt="n", yaxt="n")
par(oldpar)

## S-Curve
oldpar <- par(mfrow=c(4, 5))
T3D(SC, title="S-Curve")
for (n in 1:18) plot(Tapkee(SC, method=TM[n]), col=COL, pch=20, main=TM[n],
 xlab="", ylab="", xaxt="n", yaxt="n")
par(oldpar)

## Helix
oldpar <- par(mfrow=c(4, 5))
T3D(HX, title="Helix")
for (n in 1:18) plot(Tapkee(HX, method=TM[n]), col=COL, pch=20,
 main=TM[n], xlab="", ylab="", xaxt="n", yaxt="n")
par(oldpar)

## Severed Sphere
oldpar <- par(mfrow=c(4, 5))
T3D(SS, title="Severed Sphere", col=rainbow(nrow(SS)))
for (n in 1:18) plot(Tapkee(SS, method=TM[n]), col=rainbow(nrow(SS)), pch=20,
 main=TM[n], xlab="", ylab="", xaxt="n", yaxt="n")
par(oldpar)