l_scale3D scales its argument in a variety of ways
used for 3D visualization.
Usage
l_scale3D(x, center = TRUE, method = c("box", "sphere"))Arguments
- x
the matrix or data.frame whose columns are to be scaled. Any
NAentries will be preserved but ignored in calculations.xmust have exactly 3 columns formethod = "sphere".- center
either a logical value or numeric-alike vector of length equal to the number of columns of
x, where ‘numeric-alike’ means thatas.numeric(.)will be applied successfully ifis.numeric(.)is not true.- method
the scaling method to use. If
method = "box"(the default) then the columns are scaled to have equal ranges and, whencenter = TRUE, to be centred by the average of the min and max; Ifmethod = "sphere"thenxmust be three dimensional. For sphering, on each of the original 3 dimensionsxis first centred (mean centred whencenter = TRUE) and scaled to equal standard deviation on. The V matrix of the singular value decomposition (svd) is applied to the right resulting in uncorrelated variables. Coordinates are then divided by (non-zero as tested by!all.equal(0, .)) singular values. Ifxcontains noNAs, the resulting coordinates are simply the U matrix of the svd.
Value
a data.frame whose columns are centred and scaled according to
the given arguments. For method = "sphere"), the three variable names are
x1, x2, and x3.
Examples
##### Iris data
#
# All variables (including Species as a factor)
result_box <- l_scale3D(iris)
head(result_box, n = 3)
#> Sepal.Length Sepal.Width Petal.Length Petal.Width Species
#> 1 -0.2777778 0.12500000 -0.4322034 -0.4583333 -0.5
#> 2 -0.3333333 -0.08333333 -0.4322034 -0.4583333 -0.5
#> 3 -0.3888889 0.00000000 -0.4491525 -0.4583333 -0.5
apply(result_box, 2, FUN = range)
#> Sepal.Length Sepal.Width Petal.Length Petal.Width Species
#> [1,] -0.5 -0.5 -0.5 -0.5 -0.5
#> [2,] 0.5 0.5 0.5 0.5 0.5
# Note mean is not zero.
apply(result_box, 2, FUN = mean)
#> Sepal.Length Sepal.Width Petal.Length Petal.Width Species
#> -0.07129630 -0.05944444 -0.03254237 -0.04194444 0.00000000
# Sphering only on 3D data.
result_sphere <- l_scale3D(iris[, 1:3], method = "sphere")
head(result_sphere, n = 3)
#> x1 x2 x3
#> 1 -0.10665359 -0.03635358 -0.039744517
#> 2 -0.09153699 0.06124692 -0.080344512
#> 3 -0.11208702 0.03517574 -0.009633385
apply(result_sphere, 2, FUN = range)
#> x1 x2 x3
#> [1,] -0.1423020 -0.2338372 -0.1627308
#> [2,] 0.1737036 0.2262592 0.2203348
# Note mean is numerically zero.
apply(result_sphere, 2, FUN = mean)
#> x1 x2 x3
#> -1.854727e-17 3.437837e-18 1.035964e-16
# With NAs
x <- iris
x[c(1, 3), 1] <- NA
x[2, 3] <- NA
result_box <- l_scale3D(x)
head(result_box, n = 5)
#> Sepal.Length Sepal.Width Petal.Length Petal.Width Species
#> 1 NA 0.12500000 -0.4322034 -0.4583333 -0.5
#> 2 -0.3333333 -0.08333333 NA -0.4583333 -0.5
#> 3 NA 0.00000000 -0.4491525 -0.4583333 -0.5
#> 4 -0.4166667 -0.04166667 -0.4152542 -0.4583333 -0.5
#> 5 -0.3055556 0.16666667 -0.4322034 -0.4583333 -0.5
apply(result_box, 2, FUN = function(x) {range(x, na.rm = TRUE)})
#> Sepal.Length Sepal.Width Petal.Length Petal.Width Species
#> [1,] -0.5 -0.5 -0.5 -0.5 -0.5
#> [2,] 0.5 0.5 0.5 0.5 0.5
# Sphering only on 3D data.
result_sphere <- l_scale3D(x[, 1:3], method = "sphere")
# Rows having had any NA are all NA after sphering.
head(result_sphere, n = 5)
#> x1 x2 x3
#> 1 NA NA NA
#> 2 NA NA NA
#> 3 NA NA NA
#> 4 -0.1097604 -0.05847447 0.021141688
#> 5 -0.1188225 0.04803699 0.001622979
# Note with NAs mean is no longer numerically zero.
# because centring was based on all non-NAs in each column
apply(result_sphere, 2, FUN = function(x) {mean(x, na.rm = TRUE)})
#> x1 x2 x3
#> 0.0012171739 -0.0002142758 0.0019700846