ggplot2 Customization Exercises in R: 17 Theme & Scale Practice Problems
Seventeen runnable exercises that drill the four customization layers in ggplot2: theme presets, targeted theme() element overrides, colour and axis scales, and legend or label fine-tuning. Every problem ships with a hidden solution, an expected result, and an explanation of why the chosen approach is the idiomatic one.
Section 1. Theme presets and quick swaps (3 problems)
Exercise 1.1: Apply theme_minimal with a larger base font
Task: Build a scatter plot of mpg versus wt from mtcars, coloured by factor(cyl). Apply theme_minimal() with a base font size of 14 so the chart reads well in a slide deck. Save the finished plot object to ex_1_1.
Expected result:
#> Scatter of weight vs mpg, points coloured by cylinder count.
#> White panel, no grey background, axis labels and tick text noticeably
#> larger than ggplot2's 11-point default; faint grid lines remain.Difficulty: Beginner
Themes carry a single dial that rescales every text element at once - reach for that rather than restyling axis labels one by one.
Add theme_minimal() to the plot and pass it base_size = 14.
Click to reveal solution
Explanation: theme_minimal() strips the grey panel and axis lines but keeps faint grid lines, which works well for slides and reports. Passing base_size = 14 rescales every text element at once, so axis titles, tick labels, and legend text all grow proportionally; setting fonts piece by piece in theme() would mean four or five separate element_text() calls. Use base_family the same way to swap fonts globally.
Exercise 1.2: Compare theme_classic and theme_bw side by side
Task: Create two plot objects from the iris dataset showing Sepal.Length versus Petal.Length coloured by Species. Apply theme_classic() to the first and theme_bw() to the second so a reviewer can compare line-only versus boxed presentations. Save them to ex_1_2_classic and ex_1_2_bw, then save the list list(classic = ex_1_2_classic, bw = ex_1_2_bw) to ex_1_2.
Expected result:
#> ex_1_2$classic: scatter with axis lines only, no panel border, no grid.
#> ex_1_2$bw: scatter inside a black rectangular border with light grid.
#> Both share the iris colour mapping and identical data layer.Difficulty: Beginner
Build the shared data layer once so the two plots differ only in the surrounding style you add afterwards.
Save the common ggplot to a base object, add theme_classic() and theme_bw() to it separately, then collect both in a named list().
Click to reveal solution
Explanation: Saving the shared layer to base and adding the theme afterwards is the cleanest way to compare presets without retyping the geom. theme_classic() removes the panel border and gridlines for a publication look; theme_bw() keeps the black border plus a soft grid, which suits exploratory work where you still need to read off values. Both inherit from theme_grey() and override only the elements they change.
Exercise 1.3: Set a base font family and base line size globally
Task: Take the same mpg scatter from Exercise 1.1 and apply theme_light() with base_size = 12, base_family = "sans", and base_line_size = 0.4. The thinner baseline width should make axis lines and ticks visibly lighter than the default. Save the finished plot to ex_1_3.
Expected result:
#> Light grey panel scatter with thinner-than-default axis ticks and
#> sans-family text. No code-level errors and no warning about missing fonts.Difficulty: Intermediate
Every built-in theme exposes proportional dials for fonts, lines, and rectangles, so thinner ticks come from a line dial rather than from styling ticks directly.
Call theme_light() with base_size = 12, base_family = "sans", and base_line_size = 0.4.
Click to reveal solution
Explanation: Every built-in theme accepts base_line_size and base_rect_size, which scale all line and rectangle elements proportionally. This is the right knob when you want every axis tick, panel border, and grid line uniformly thinner; reaching for individual element_line(linewidth = 0.4) calls is verbose and forgets to update grid lines. "sans" always resolves on Windows, macOS, and Linux, so it is a safer default than naming Helvetica or Arial directly.
Section 2. Targeted theme() element overrides (3 problems)
Exercise 2.1: Rotate x-axis tick labels for crowded categorical axes
Task: A retailer compiled price-class counts from the diamonds dataset by cut and clarity, and the long clarity labels overlap when plotted unrotated. Build a bar chart of count(diamonds, clarity) with clarity on x and n on y, then rotate the x-axis tick labels 45 degrees with right-justified anchoring. Save to ex_2_1.
Expected result:
#> Bar chart, 8 bars (I1, SI2, SI1, VS2, VS1, VVS2, VVS1, IF) on x.
#> Tick labels printed at 45 degrees, ending flush against the tick.
#> Counts on y range from roughly 700 to 13000.Difficulty: Intermediate
Rotated tick text pivots around its anchor point, so the angle alone is not enough - you also need to control where the text is justified.
Inside theme(), set axis.text.x = element_text(angle = 45, hjust = 1).
Click to reveal solution
Explanation: axis.text.x inherits from axis.text, so overriding it touches only the bottom axis and leaves y tick text alone. hjust = 1 is the trick people miss: rotated text rotates around its anchor, and without right-justification the labels float away from their ticks. For more than ten categories or words longer than six characters, prefer coord_flip() or scale_x_discrete(guide = guide_axis(n.dodge = 2)) rather than steeper angles.
Exercise 2.2: Move the legend, restyle the legend background, and add a panel border
Task: Style a geom_point() of mpg versus wt from mtcars, coloured by factor(cyl). Move the legend to the bottom of the plot, give the legend background a pale grey fill with no border, and draw a thin black border around the panel. Save the finished plot to ex_2_2.
Expected result:
#> Scatter inside a thin black panel border. Legend sits below the plot,
#> oriented horizontally, on a pale grey rectangular background with no
#> visible legend border.Difficulty: Intermediate
Legend placement, the legend's rectangular backdrop, and the panel frame are three independent settings, each targeting a different region of the plot.
In theme(), set legend.position = "bottom", legend.background = element_rect(fill = "grey95", colour = NA), and panel.border = element_rect(fill = NA, colour = "black").
Click to reveal solution
Explanation: element_rect() is the right element for any rectangular region: legend background, panel border, plot background, facet strips. fill = NA is the only way to keep the panel border line without filling the panel itself; passing fill = "white" would override your colour scale's defaults. Setting colour = NA on the legend background suppresses its outline cleanly without resorting to element_blank(), which would also remove the fill.
Exercise 2.3: Hide the minor grid lines and lighten the major grid
Task: Build a line chart of economics$unemploy over economics$date. Hide all minor grid lines using element_blank(), and recolour the major grid lines to a very light grey (grey90) with linewidth = 0.3. Save the plot to ex_2_3.
Expected result:
#> Time-series line of US unemployment, 1967-2015.
#> No minor grid lines visible.
#> Major grid lines present but barely visible, in pale grey.Difficulty: Intermediate
Major and minor grid lines are separate leaf-level settings, so you can erase one while only recolouring the other.
In theme(), set panel.grid.minor = element_blank() and panel.grid.major = element_line(colour = "grey90", linewidth = 0.3).
Click to reveal solution
Explanation: panel.grid.minor and panel.grid.major inherit from panel.grid, so overriding them at the leaf level lets you keep major grid hints while killing the minor ones. element_blank() is the canonical "remove this" element; setting colour = NA would leave an invisible line that still consumes layout space. Pale major grid lines are a common journal style: they orient the eye without competing with the data layer.
Section 3. Colour and fill scales (3 problems)
Exercise 3.1: Apply a manual three-colour palette to a categorical fill
Task: Plot the count(diamonds, cut) distribution as a bar chart with cut on x and n on y, filling by cut. Apply a manual five-colour palette using scale_fill_manual() with the values c("Fair" = "#7f7f7f", "Good" = "#bcbd22", "Very Good" = "#17becf", "Premium" = "#e377c2", "Ideal" = "#1f77b4"). Save to ex_3_1.
Expected result:
#> Five vertical bars labelled Fair, Good, Very Good, Premium, Ideal.
#> Bars filled with grey, olive, teal, pink, blue in that left-to-right order.
#> Y axis spans 0 to ~22000.Difficulty: Intermediate
A manual palette is safest when each colour is tied to a named level rather than left to positional order.
Use scale_fill_manual(values = ...) with a vector named by each cut level.
Click to reveal solution
Explanation: Naming the palette vector by level (rather than passing an unnamed vector of five colours) is defensive: the colour for "Ideal" stays correct even if the factor levels get reordered upstream. scale_fill_manual() overrides the default discrete hue cycle and triggers an automatic legend keyed off the same names. If a level is missing from values, ggplot2 will raise a clear error instead of silently recycling a colour.
Exercise 3.2: Use scale_color_brewer with a sequential palette for an ordered factor
Task: Plot price versus carat from diamonds, coloured by clarity. Because clarity is an ordered factor running from worst (I1) to best (IF), pick scale_colour_brewer(palette = "YlOrRd", direction = -1) so the worst clarity reads as deep red and the best as pale yellow. Save the finished plot to ex_3_2.
Expected result:
#> Scatter of carat (x) vs price (y) on a busy plot.
#> Eight colours from deep red (I1) to pale yellow (IF) along the legend,
#> reflecting the ordered nature of clarity from worst to best.Difficulty: Intermediate
An ordered factor calls for a ramp that encodes magnitude, and the ramp can be flipped instead of hunting for a reversed palette name.
Use scale_colour_brewer(palette = "YlOrRd", direction = -1).
Click to reveal solution
Explanation: direction = -1 flips a ColorBrewer ramp without you having to look up the reversed palette name. Sequential ramps like YlOrRd encode magnitude order, which matches an ordered factor like clarity; using Set1 (a qualitative palette) would imply the categories are unordered and waste the hue information. alpha = 0.4 is essential here because there are 53,940 points; without it the lower clarities completely overplot the higher ones.
Exercise 3.3: Map a continuous variable to scale_fill_viridis_c with reversed direction
Task: Compute the count of diamonds per (cut, color) combination using count(diamonds, cut, color), then draw a heatmap with cut on x, color on y, and the count n mapped to fill. Apply scale_fill_viridis_c(option = "magma", direction = -1) so the highest counts read as bright yellow and the lowest as deep purple. Save the plot to ex_3_3.
Expected result:
#> 5 x 7 tile grid (5 cuts on x, 7 colors D-J on y).
#> Tiles coloured on the magma ramp, brightest yellow tiles in the
#> Ideal column for mid-range colors (E, F, G), darkest purple in
#> Fair-color-J corner.Difficulty: Intermediate
A continuous fill needs the continuous variant of the palette function, and the bright end of the ramp can be flipped onto the high values.
Apply scale_fill_viridis_c(option = "magma", direction = -1) to a geom_tile().
Click to reveal solution
Explanation: The _c suffix marks viridis as the continuous variant; _d would expect a discrete fill and error here. Magma works well on either light or dark backgrounds and is colour-blind safe like all viridis options. direction = -1 makes the brightest end of the ramp encode the largest values, which inverts the default and matches the convention that "more = lighter" in heatmaps used in printed reports.
Section 4. Continuous and discrete axes (3 problems)
Exercise 4.1: Format y-axis tick labels as US dollars with thousand separators
Task: A finance team wants the price axis on a diamonds carat-vs-price scatter to read as currency. Plot price versus carat and apply scale_y_continuous(labels = label_dollar()) from the scales package so a tick at 5000 prints as $5,000. Save the plot to ex_4_1.
Expected result:
#> Carat vs price scatter.
#> Y-axis tick labels read as $0, $5,000, $10,000, $15,000 (or similar).
#> No errors about missing labels, no warning about formatter.Difficulty: Intermediate
Axis tick text is reformatted by handing the scale a formatting function rather than editing label strings by hand.
Pass labels = label_dollar() to scale_y_continuous().
Click to reveal solution
Explanation: label_dollar() is the modern replacement for the old dollar_format() and ships with sensible defaults: dollar prefix, comma thousands separator, no decimal places. Passing it as labels = label_dollar() (with parens) is intentional, the scale needs the formatting function, not its name. For other currencies, swap in prefix = "EUR " or suffix = " kr"; for percentages, the analogous helper is label_percent().
Exercise 4.2: Set custom breaks, limits, and a log10 transformation on the x-axis
Task: A pricing analyst wants to inspect the long tail of diamond prices on a log scale. Plot price (x) versus carat (y) from diamonds, applying scale_x_log10(limits = c(300, 20000), breaks = c(500, 1000, 2500, 5000, 10000), labels = label_dollar()) so the x-axis reads as currency on a log10 grid with the listed break values. Save to ex_4_2.
Expected result:
#> Scatter with log-spaced x-axis showing $500, $1,000, $2,500, $5,000, $10,000.
#> Points span from carat ~0.2 (low) to ~5 (high) on the y-axis.
#> A few points outside the [300, 20000] price window are dropped (with
#> a warning about removed rows).Difficulty: Advanced
A transform, the visible break values, and the data window can all be configured inside the single scale that governs that axis.
Use scale_x_log10() with limits, breaks, and labels = label_dollar() arguments.
Click to reveal solution
Explanation: Combining breaks, labels, and a transform inside one scale_x_log10() call is cleaner than chaining separate scale functions, which would conflict. limits here clips data, so a few extreme rows drop and ggplot warns; if you only want to zoom the view without dropping data, use coord_cartesian(xlim = ...) instead. label_dollar() operates on the original units, not the log-transformed ones, which is exactly what readers expect.
Exercise 4.3: Reorder a discrete x-axis by the median y value
Task: Plot a boxplot of mpg per class from the mpg dataset, but reorder the x-axis so classes are sorted by ascending median highway mileage rather than alphabetically. Use aes(x = reorder(class, hwy, FUN = median), y = hwy), then relabel the x-axis as "Vehicle class (sorted by median highway mpg)" via labs(). Save to ex_4_3.
Expected result:
#> Boxplot per vehicle class, x ordered left-to-right from lowest to
#> highest median hwy (e.g. pickup, suv, minivan, midsize, ..., compact, subcompact).
#> X-axis title reads "Vehicle class (sorted by median highway mpg)".Difficulty: Advanced
A discrete axis follows its factor's level order, so the sorting must happen on the variable mapped to x, driven by a summary of y.
Map x with reorder(class, hwy, FUN = median) and rename the axis with labs().
Click to reveal solution
Explanation: reorder(x, y, FUN) is the base R way to make a factor whose level order is driven by a summary of another variable; ggplot2 reads that order off the factor and uses it for the discrete axis. The forcats package offers fct_reorder() with the same semantics if you prefer the tidyverse spelling. Without reorder(), the x-axis sorts factor levels alphabetically, which is rarely what a reader wants for ordinal comparisons.
Section 5. Legends, labels, and annotations (3 problems)
Exercise 5.1: Override the colour legend title and remove the size legend
Task: Plot a geom_point() of mpg versus wt from mtcars, mapping colour = factor(cyl) and size = hp. Use labs(colour = "Cylinders") to rename the colour legend, and use guides(size = "none") to suppress the hp size legend so the chart only shows one legend entry. Save the final plot to ex_5_1.
Expected result:
#> Scatter with point sizes that vary by hp (visible) but only one legend
#> on the right titled "Cylinders" with three entries (4, 6, 8).
#> No "hp" or "size" legend block visible.Difficulty: Intermediate
Renaming a legend and hiding a different one are separate jobs - one touches a title, the other turns a guide off without unmapping the aesthetic.
Use labs(colour = "Cylinders") to rename and guides(size = "none") to suppress the size legend.
Click to reveal solution
Explanation: labs() is the convenience wrapper for axis and legend titles, and renaming the legend is just labs(colour = "...") matched to whichever aesthetic you mapped. guides(size = "none") is the canonical way to hide a single legend without disabling the underlying mapping; setting theme(legend.position = "none") would hide every legend on the plot. Use guide_legend(override.aes = list(size = 4)) when you need to keep a legend but want consistent symbol sizes.
Exercise 5.2: Add a styled title, subtitle, and caption with custom typography
Task: Build a geom_point() of mpg versus wt from mtcars. Use labs() to set a title, subtitle, and caption, then style them inside theme(): title bold and 16pt, subtitle italic and grey40, caption right-aligned and 9pt. Save the plot to ex_5_2.
Expected result:
#> Scatter with three text rows above and below the panel.
#> Title in bold 16pt, subtitle in italic grey, caption right-aligned in
#> small grey text below the panel. The plot itself is unchanged.Difficulty: Advanced
Title, subtitle, and caption are three independent text elements, so each one is styled on its own to build the typographic hierarchy.
Set the text with labs(), then style plot.title, plot.subtitle, and plot.caption using element_text() inside theme().
Click to reveal solution
Explanation: plot.title, plot.subtitle, and plot.caption are independent elements; styling each separately gives the bold/italic/small typographic hierarchy expected in a published chart. The caption defaults to right alignment in newer ggplot2, but setting hjust = 1 makes the intent explicit and survives a future default change. Use theme(plot.title.position = "plot") if you want titles to align with the full plot edge rather than the panel edge.
Exercise 5.3: Annotate a single highlighted point with an arrow and label
Task: A code reviewer wants the most fuel-efficient car in mtcars flagged on a scatter of mpg versus wt. Identify the row with the maximum mpg (Toyota Corolla, mpg = 33.9, wt = 1.835), then build a scatter that adds an annotate("text", ...) label reading the car's name and an annotate("segment", ...) arrow pointing to the row. Save the plot to ex_5_3.
Expected result:
#> Scatter of wt (x) vs mpg (y).
#> A short arrow ends near the top-left point; the text "Toyota Corolla"
#> sits above-and-right of the arrow's start.Difficulty: Advanced
Marking one fixed observation calls for layers whose coordinates are constants, not values pulled from the plotted data frame.
Add annotate("segment", ..., arrow = arrow(...)) for the arrow and annotate("text", ..., label = ...) for the label.
Click to reveal solution
Explanation: annotate() adds geoms whose values are constants, not pulled from a data frame, so they do not pollute the colour or fill scale. arrow(length = unit(0.2, "cm")) from the grid package controls arrowhead size, and unit() is loaded by ggplot2 itself. For label collision avoidance with many highlighted points, switch to ggrepel::geom_text_repel(), but annotate() is the cleanest tool when you want to flag exactly one observation.
Section 6. Composing a reusable house style (3 problems)
Exercise 6.1: Define a custom theme function and apply it to two different plots
Task: A reporting team wants every chart in their weekly digest to share the same look. Write a function theme_digest() that wraps theme_minimal(base_size = 12) and then layers a theme() block setting bold 14pt title, grey40 axis text, and panel.grid.minor = element_blank(). Apply it to a scatter of mpg versus wt from mtcars (save as p_scatter) and to a column chart of count(diamonds, cut) (save as p_bar). Save list(scatter = p_scatter, bar = p_bar) to ex_6_1.
Expected result:
#> ex_6_1$scatter and ex_6_1$bar render with identical typography:
#> bold 14pt title, grey40 axis text, no minor grid.
#> Both inherit theme_minimal otherwise. Function reusable on any plot.Difficulty: Advanced
Bundling repeated styling into one reusable definition keeps the house look in a single place to update.
Write a function that returns theme_minimal(base_size = 12) + theme(...), then add theme_digest() to both plots.
Click to reveal solution
Explanation: Wrapping the boilerplate in a function gives a single place to update the house style; revising the title size touches one definition rather than every plot. theme_minimal() + theme(...) returns a single theme object, which is what + adds to the plot. The base_size parameter lets the same function produce slide-deck-sized variants without duplicating code, and you can extend the pattern with scale_* defaults via scale_colour_discrete() overrides.
Exercise 6.2: Combine a custom theme with a custom default colour scale
Task: Extend Exercise 6.1 by writing a second helper, digest_palette, that returns a named character vector of three brand colours: navy "#1f3b73", gold "#d4a017", and crimson "#a23b3b". Build a geom_point() of mpg versus wt from mtcars coloured by factor(cyl) and apply both theme_digest() and scale_colour_manual(values = digest_palette). Save the finished plot to ex_6_2.
Expected result:
#> Scatter with three colours, navy / gold / crimson, mapped to cyl 4, 6, 8.
#> Same typography as ex_6_1 (bold title, grey40 axis text, no minor grid).
#> Legend titled "factor(cyl)" with three swatches in brand colours.Difficulty: Advanced
Typography and brand colour are separate concerns, so keep the theme helper and the palette as two distinct objects.
Define digest_palette as a vector named by cyl level, then add scale_colour_manual(values = digest_palette) alongside theme_digest().
Click to reveal solution
Explanation: Naming the palette by factor level ("4", "6", "8") locks the mapping so cylinder 4 always renders navy regardless of factor sort order. Splitting palette and theme into two separate helpers keeps each concern focused: typography lives in theme_digest(), colour lives in digest_palette. For full brand kits, expose a scale_colour_digest() convenience function that wraps scale_colour_manual(values = digest_palette, ...) so analysts never type the palette name directly.
Exercise 6.3: Save a plot to disk at a specific size and resolution for print
Task: Take the ex_6_2 plot from Exercise 6.2 and save it to disk as digest_chart.png at 6 inches wide, 4 inches tall, and 300 dpi using ggsave(). Capture the absolute file path returned by normalizePath("digest_chart.png") (after saving) and store it in ex_6_3. The check is whether the saved file exists at that path with non-zero size.
Expected result:
#> ex_6_3 contains an absolute path string ending in "digest_chart.png".
#> file.exists(ex_6_3) is TRUE.
#> file.info(ex_6_3)$size is greater than 1000 bytes (a real PNG, not empty).Difficulty: Intermediate
Reproducible output size comes from stating dimensions and resolution explicitly rather than relying on the current graphics device.
Call ggsave() with width = 6, height = 4, units = "in", and dpi = 300, then capture the path with normalizePath().
Click to reveal solution
Explanation: ggsave() infers the output device from the file extension, so .png triggers PNG, .pdf triggers PDF, no extra arguments needed. Specifying width, height, and units = "in" together with dpi = 300 produces a 1800x1200 raster suitable for print; the default device size is whatever your current graphics window happens to be, which is rarely reproducible. For vector output, switch to .pdf or .svg and drop the dpi argument since vectors are resolution-independent.
What to do next
- Drill the foundational theme grammar in ggplot2 Themes in R, the parent post for this hub.
- Practice the colour scales side of customization with ggplot2 Color Scales Exercises in R.
- Build out the underlying chart skills first via ggplot2 Exercises in R.
- Step into multi-panel layout work with ggplot2 Facets Exercises in R.
r-statistics.co · Verifiable credential · Public URL
This document certifies mastery of
ggplot2 Customization (17 problems) Mastery
Every certificate has a public verification URL that proves the holder passed the assessment. Anyone with the link can confirm the recipient and date.
164 learners have earned this certificate