Skip to contents

Contrast weight heatmap

Source: R/plot_heatmaps.R
plot_contrasts.event_model.Rd

Shows every contrast defined on an event_model as a row of weights over all design-matrix columns, in design order and grouped by term, so it lines up column-for-column with design_map(). Weights use the package's diverging palette, symmetric about zero up to the largest absolute weight across all contrasts (stated in the legend), and every weight is printed: zero weights sit at the neutral midpoint with a faint "0", so exclusion reads as an explicit weight rather than a missing cell.

Usage

# S3 method for class 'event_model'
plot_contrasts(
  x,
  absolute_limits = FALSE,
  rotate_x_text = TRUE,
  scale_mode = c("auto", "diverging", "one_sided"),
  coord_fixed = FALSE,
  annotate = NULL,
  title = "Contrast weights",
  subtitle = NULL,
  ...
)

Arguments

x

An event_model with contrasts defined in its terms.

absolute_limits

Logical; if TRUE, fix the fill scale at [-1, 1]. Otherwise it is symmetric about 0 up to the largest abs(weight).

rotate_x_text

Logical; angle column labels when they would overlap.

scale_mode

"auto" or "diverging" (both diverging and symmetric about 0), or "one_sided" (neutral-to-warm magnitude scale from 0, for non-negative weights).

coord_fixed

Logical; if TRUE, keep cells square. The default (FALSE) uses shallow rows of fixed height.

annotate

Logical or NULL; print weights in the cells. NULL (default) prints them when there are at most 40 columns.

title, subtitle

Plot title and subtitle.

...

Passed to ggplot2::geom_tile().

Value

A ggplot object.

Details

Two columns in the right margin check each row. "sum w" is the sum of the weights; non-zero sums are shown in a warning colour. The SE multiplier is sqrt(c' (X'X)^+ c), with X the design matrix plus one intercept per run: the standard error of the contrast estimate per unit of noise standard deviation, given this design (lower is better). Contrasts that are not estimable from the design are marked "not est.". F-contrasts appear as one row per component, numbered under the contrast name, each with its own values.

See also

contrast_weights(), design_map()

Examples

des <- data.frame(
  onset = c(0, 10, 20, 30, 40, 50),
  run = 1,
  cond = factor(c("A", "B", "C", "A", "B", "C"))
)
sframe <- fmrihrf::sampling_frame(blocklens = 60, TR = 1)
cset <- contrast_set(
  A_vs_B = pair_contrast(~ cond == "A", ~ cond == "B", name = "A_vs_B"),
  B_vs_C = pair_contrast(~ cond == "B", ~ cond == "C", name = "B_vs_C")
)
emod <- event_model(onset ~ hrf(cond, contrasts = cset),
                    data = des, block = ~run, sampling_frame = sframe)
plot_contrasts(emod)