svd S4 Class

S4 class for deflation SVD/PCA results.

Computes scores for new samples by projecting newdata onto the right singular vectors. Equivalent to PCA "out-of-sample prediction".

Usage

# S4 method for class 'svd,ANY,ANY,ANY'
x[i]

# S4 method for class 'svd'
head(x, n = getOption("digits"), ...)

# S4 method for class 'svd'
show(object)

# S4 method for class 'svd'
dim(x)

reconstruct(object, ...)

# S4 method for class 'svd'
reconstruct(object, ...)

# S4 method for class 'svd'
predict(object, newdata, ...)

variance_explained(object, ...)

# S4 method for class 'svd'
variance_explained(object, ...)

Arguments

x

object of class svd

i

indices for subsetting factors

n

number of rows/columns to show

...

Ignored.

object

An svd object

newdata

A numeric matrix of new samples (rows = samples, columns = features). Must have the same number of features as the original data (ncol(newdata) == nrow(object@v)).

Value

A subsetted svd object containing only the selected factors.

Invisibly returns the svd object x.

Invisibly returns the svd object.

Integer vector of length 3: c(m, n, k) where m is the number of rows, n the number of columns, and k the rank.

Dense matrix: \(U \cdot diag(d) \cdot V'\) (plus row means if centered)

A numeric matrix of scores with nrow(newdata) rows and length(object@d) columns (i.e., same \(k\) as the model). Each row is the projection of one new sample onto the singular space.

Numeric vector of proportion of variance explained by each factor

Slots

u

left singular vectors (scores), m x k matrix

d

singular values, numeric vector of length k

v

right singular vectors (loadings), n x k matrix

misc

list containing metadata: centered, row_means, test_loss, iters_per_factor, wall_time_ms, auto_rank

See also

svd, pca, reconstruct, variance_explained

Examples

# \donttest{
A <- matrix(rnorm(200), 20, 10)
s <- RcppML::svd(A, k = 3)
Ahat <- reconstruct(s)
dim(Ahat)
#> [1] 20 10
# }
# \donttest{
A <- matrix(rnorm(200), 20, 10)
s <- RcppML::svd(A, k = 3)
variance_explained(s)
#> [1] 0.2249887 0.1938869 0.1546420
# }