create.polygonplot bugs when `groups = NULL`

[jarbet]

Example data

suppressPackageStartupMessages(library(BoutrosLab.plotting.general));
suppressPackageStartupMessages(library(mgcv));

alpha <- 0.05;
zcrit <- qnorm(1 - alpha / 2);
fit <- gam(
    mpg ~ s(hp),
    data = mtcars
    );

pred.grid <- data.frame(
    hp = seq(50, 300, by = 0.01)
    );
pred <- data.frame(predict(
    object = fit,
    newdata = pred.grid,
    type = 'response',
    se.fit = TRUE
    ));
pred$lower <- pred$fit - zcrit * pred$se.fit;
pred$upper <- pred$fit + zcrit * pred$se.fit;
pred$hp <- pred.grid$hp;
head(pred)
#>        fit   se.fit    lower    upper    hp
#> 1 30.22140 1.511997 27.25794 33.18486 50.00
#> 2 30.21997 1.511678 27.25713 33.18280 50.01
#> 3 30.21854 1.511359 27.25633 33.18075 50.02
#> 4 30.21711 1.511041 27.25552 33.17869 50.03
#> 5 30.21568 1.510722 27.25472 33.17664 50.04
#> 6 30.21425 1.510404 27.25391 33.17458 50.05

xat <- seq(50, 300, by = 50);
yat <- seq(0, 40, by = 10);
col <- default.colours(1);

Plot doesnt work for default xat/xlimits/yat/ylimits

create.polygonplot(
    formula = NA ~ hp,
    data = pred,
    min = pred$lower,
    max = pred$upper,
    col = col,
    add.median = TRUE,
    median = pred$est
    );

border.col/alpha do not work when groups = NULL

Related to #24

create.polygonplot(
    formula = NA ~ hp,
    data = pred,
    min = pred$lower,
    max = pred$upper,
    ylimits = c(min(yat), max(yat)),
    yat = yat,
    xlimits = c(min(xat), max(xat)),
    xat = xat,
    col = col,
    alpha = 0.3, # doesnt work
    grid.col = 'transparent',
    add.border = TRUE,
    border.col = 'transparent', # doesnt work
    add.median = TRUE,
    median = pred$fit,
    median.col = 'black',
    main = 'border.col = \'transparent\', alpha = 0.3\' do not work when groups = NULL',
    main.cex = 0.9
    );

Solution

For now, the solution is to provide a madeup constant group variable:

create.polygonplot(
    formula = NA ~ hp,
    data = pred,
    min = pred$lower,
    max = pred$upper,
    ylimits = c(min(yat), max(yat)),
    yat = yat,
    xlimits = c(min(xat), max(xat)),
    xat = xat,
    col = col,
    alpha = 0.3, # doesnt work
    grid.col = 'transparent',
    add.border = TRUE,
    border.col = 'transparent',
    add.median = TRUE,
    median = pred$fit,
    median.col = 'black',
    main = 'border.col = \'transparent\' and alpha = 0.3\' work with madeup groups',
    main.cex = 1,
    groups = rep(1, nrow(pred)) # madeup constant groups
    );

^{Created on 2025-08-05 with reprex v2.1.1}

[jarbet]

@alkaZeltser : it seems you fixed all of these bugs in your plotting example here(code), can you please share your solution?

[alkaZeltser]

@jarbet if the bug is related to the status of the grouping variable, I think all my examples used groups, which would explain why I didn't have any issues with transparency

[jarbet]

@jarbet if the bug is related to the status of the grouping variable, I think all my examples used groups, which would explain why I didn't have any issues with transparency

Yep that's right! I modified my example code above and put a temporary solution we can all use (supplying madeup constant groups fixes everything). Another important issue is the user will likely always need to supply xat/yat/xlimits/ylimits in order to get the plot to work.

Eventually we can fix this bug such that the user will not need to supply madeup groups if there are no groups. This should be an easy fix, I imagine we can add something like:

if (is.null(groups)) {
    groups <- rep(1, nrow(data));
    }