flash_plot_scree: Create a scree plot for a flash fit

In willwerscheid/flashier: Empirical Bayes Matrix Factorization

Create a scree plot for a flash fit

Description

A scree plot is a line plot showing the proportion of variance explained by each factor/loadings pair in a flash fit. Note that since EBMF does not require factors and loadings to be orthogonal, "PVE" should be interpreted loosely: for example, the total proportion of variance explained could be larger than 1.

Usage

flash_plot_scree(
  fl,
  order_by_pve = FALSE,
  kset = NULL,
  labels = FALSE,
  label_size = 3,
  max_overlaps = Inf
)

Arguments

fl	An object inheriting from class flash.
order_by_pve	If order_by_pve = TRUE, then factor/loadings pairs will be ordered according to proportion of variance explained, from highest to lowest. (By default, they are plotted in the same order as kset; or, if kset is NULL, then they are plotted in the same order as they are found in fl.)
kset	A vector of integers specifying the factor/loadings pairs to be plotted. If order_by_pve = FALSE, then kset also specifies the order in which they are to be plotted.
labels	Whether to label the points in the scree plot with the indices of the factor/loading pairs they correspond to. Labels appear as "k1", "k2", "k3", etc.
label_size	The size of the label text (in millimeters).
max_overlaps	A (nonnegative) integer. For each text label, the number of overlaps with other text labels or other data points are counted, and the text label is excluded if it has too many overlaps.

Details

Unlike scree plots for PCA, a scree plot for a flash fit is not in general monotonically decreasing. To ensure a monotonically decreasing scree plot, set order_by_pve = TRUE. Note, however, that if this is done then the numbers on the x-axis will no longer match the indices of the components in the flash fit. This can also be true if argument kset has been specified. Thus one should consider setting labels = TRUE when order_by_pve = TRUE or when kset has been specified.

Value

A ggplot object.

Examples

data(gtex)
fl <- flash(gtex, greedy_Kmax = 4L, backfit = FALSE)
flash_plot_scree(fl)

# For the full range of labelling options provided by the ggrepel package, set
#   labels = FALSE (the default setting) and add geom_text_repel() manually:
library(ggrepel)
flash_plot_scree(fl) + geom_text_repel(min.segment.length = 0)

willwerscheid/flashier documentation built on Jan. 30, 2025, 11:27 p.m.