ggplot2 Heatmap Exercises in R: 20 Real-World Practice Problems
Twenty interactive heatmap problems covering geom_tile fundamentals, diverging and viridis palettes, hierarchical row ordering, correlation matrices with significance masking, and faceted cohort retention. Each problem ships with a hidden full solution and a why-it-works explanation.
Section 1. Foundations with geom_tile (4 problems)
Exercise 1.1: Build a first heatmap from a long tibble of weekly call volume
Task: A telco operations lead wants a heatmap of inbound call volume by day (Mon to Sun) and hour (0 to 23) from a small inline tibble called calls. Use geom_tile() with x = hour, y = day, fill = volume to produce the heatmap and save the ggplot object to ex_1_1.
Expected result:
#> A ggplot heatmap with x = hour (0..23), y = day (Mon..Sun),
#> fill = volume in default blue gradient.
#> 7 rows x 24 cols = 168 tiles, no axis titles overridden.Difficulty: Beginner
A heatmap is just one colored rectangle per row of data, so map your three columns to horizontal position, vertical position, and color.
Pass aes(x = hour, y = day, fill = volume) to ggplot() and add a tile layer.
Click to reveal solution
Explanation: geom_tile() draws one rectangle per row of the data frame, centered on (x, y). Because calls is already in long format (one row per cell), no reshaping is needed. The default continuous fill scale runs light-to-dark blue and is fine for a quick look. For a polished version you would override the palette and the categorical y-axis ordering, but the minimum heatmap is three aesthetics and one geom.
Exercise 1.2: Reshape airquality into a month-by-day Ozone heatmap
Task: The built-in airquality dataset has one row per day with Month, Day, and Ozone. Pivot it into a heatmap with Day on the x-axis, Month on the y-axis, and fill encoding Ozone level. Drop rows where Ozone is missing before plotting, then save the ggplot object to ex_1_2.
Expected result:
#> A heatmap: x = Day (1..31), y = Month (5..9 as discrete),
#> fill = Ozone (ppb). Gaps where Ozone was NA.
#> Approximately 116 tiles drawn (153 days - 37 NA rows).Difficulty: Beginner
Remove the missing readings first, then map day, month, and ozone onto the three heatmap channels, keeping the month axis categorical.
Use filter(!is.na(Ozone)) upstream and wrap the month in factor(Month) inside aes() before the tile layer.
Click to reveal solution
Explanation: Wrapping Month in factor() forces it to be discrete, which is what you want for the y-axis of a calendar-style heatmap. If you left it numeric, ggplot2 would treat the axis as continuous and draw tiles overlapping the integer gridlines. Dropping NA Ozone rows is cleaner than passing them through because the default na.value = "grey50" clashes with the blue palette. An alternative is to keep NA rows and set na.value = "white" in the scale, covered in exercise 6.2.
Exercise 1.3: Compare geom_tile and geom_raster on a 200x200 grid
Task: When the grid is large and regularly spaced, geom_raster() is much faster than geom_tile() because it draws a single rasterized image instead of N rectangles. Build a 200x200 grid of (x, y, z) where z = sin(x/10) * cos(y/10), draw it with geom_raster(), and save the plot to ex_1_3.
Expected result:
#> A smooth raster image, 200 x 200 pixels, with diagonal
#> sinusoidal banding (interference pattern). Default fill
#> scale (light = high z, dark = low z).Difficulty: Intermediate
On a large, evenly spaced grid, draw the whole surface as a single rasterized image rather than as tens of thousands of separate rectangles.
Map x, y, and z in aes() and add geom_raster() instead of the per-cell tile geom.
Click to reveal solution
Explanation: geom_raster() requires the data to lie on an evenly spaced grid; if rows are missing or spacing is irregular, it falls back to geom_tile() semantics but with no NA warning. For 40,000 cells the speedup over geom_tile() is typically 5x to 20x. A common mistake is to use geom_tile() on regular grids out of habit and then complain about render time. Rule of thumb: regular grid means raster, irregular or sparse means tile.
Exercise 1.4: Heatmap of a manually built customer churn risk matrix
Task: A risk analyst wants to see churn rate by tenure_band (rows: 0-6m, 7-12m, 13-24m, 25m+) and plan (cols: Basic, Pro, Enterprise). Build the inline tibble below, then plot a geom_tile() heatmap with churn percentage as fill. Save the ggplot object to ex_1_4.
Expected result:
#> A 4x3 heatmap. Highest churn cell: 0-6m / Basic at ~38%.
#> Lowest cell: 25m+ / Enterprise at ~4%. Blue gradient.Difficulty: Intermediate
The tibble is already one row per cell, so the heatmap is a direct mapping of plan, tenure band, and churn rate.
Build ggplot() with aes(x = plan, y = tenure_band, fill = churn_pct) and add a tile layer.
Click to reveal solution
Explanation: Once data is already in long format (one row per cell), the heatmap is one line. The interesting work is upstream: producing the long tibble. In a real churn study the tenure_band order matters and would need factor(tenure_band, levels = c("0-6m", "7-12m", "13-24m", "25m+")) to keep increasing tenure going up the axis; ggplot2 sorts character levels alphabetically by default which scrambles the bands here.
Section 2. Color scales and palettes (4 problems)
Exercise 2.1: Apply scale_fill_gradient with custom low and high colors
Task: Take the heatmap from exercise 1.1 (call volume by day and hour) and replace the default blue scale with a white-to-firebrick gradient using scale_fill_gradient(low = "white", high = "firebrick"). Save the new ggplot object to ex_2_1 so the heatmap reads visually as "white means quiet, red means busy".
Expected result:
#> Same 7x24 heatmap as ex_1_1 but with white-to-red fill scale.
#> The busiest cells appear deep red; the quietest appear near-white.Difficulty: Beginner
Replace the default color ramp with a two-color continuous scale running from a pale low end to a saturated high end.
Add scale_fill_gradient(low = "white", high = "firebrick") to the existing plot.
Click to reveal solution
Explanation: scale_fill_gradient() is the simplest two-color continuous scale. Pick a low color that fades into the chart background (often "white" or a very light grey) and a high color with enough saturation to draw the eye. Avoid pure red ("red") plus pure green; reserve red for "bad" or "high attention" semantics and prefer firebrick or red4 for a slightly muted look. For three-color (diverging) palettes use scale_fill_gradient2() (exercise 2.2).
Exercise 2.2: Diverging palette around zero for a correlation tile grid
Task: Build a small inline correlation matrix of four financial metrics and draw it as a heatmap with a diverging palette centered on zero so that positive and negative correlations are visually opposite. Use scale_fill_gradient2(low = "steelblue", mid = "white", high = "firebrick", midpoint = 0, limits = c(-1, 1)) and save the plot to ex_2_2.
Expected result:
#> A 4x4 heatmap; diagonal is deep red (corr = 1).
#> Off-diagonal: negative cells (e.g. price vs vol = -0.6) are blue,
#> positive cells (e.g. returns vs alpha = 0.4) are pink.Difficulty: Intermediate
Correlations have a meaningful zero, so the color scale should pivot at the middle with opposite hues for negative and positive values.
Add a tile layer and scale_fill_gradient2() with midpoint = 0 and limits = c(-1, 1).
Click to reveal solution
Explanation: scale_fill_gradient2() is the canonical pick for any quantity with a meaningful zero or midpoint (correlations, log-fold changes, deviations from a baseline). Setting limits = c(-1, 1) makes the legend symmetric so a correlation of 0.5 visually balances minus 0.5; without the explicit limits, ggplot2 would auto-fit to the data and break that symmetry. Avoid sequential palettes (scale_fill_gradient(), viridis) for diverging data because they hide the sign.
Exercise 2.3: Colorblind-friendly viridis for the airquality heatmap
Task: Take the airquality month-by-day Ozone heatmap structure from exercise 1.2 and apply scale_fill_viridis_c(option = "magma", direction = -1) so that high Ozone is dark and low is bright. The reversed magma palette is print-safe and colorblind-friendly. Save the ggplot object to ex_2_3.
Expected result:
#> A month-by-day heatmap with magma palette, reversed:
#> low Ozone = bright yellow, high Ozone = dark purple.
#> 116 tiles drawn (NA days dropped before plotting).Difficulty: Intermediate
Swap in a perceptually uniform, colorblind-safe palette and flip its direction so that dark marks the high end.
After the tile layer, add scale_fill_viridis_c(option = "magma", direction = -1).
Click to reveal solution
Explanation: Viridis palettes (viridis, magma, inferno, plasma, cividis) are perceptually uniform, meaning equal steps in data map to equal perceived steps in color, and they remain distinguishable in greyscale and for common colorblindness types. Using direction = -1 reverses the palette so that dark cells flag "high" rather than "low", which often reads more naturally for pollution, risk, or error magnitude. For discrete fills swap in scale_fill_viridis_d().
Exercise 2.4: Discretize a continuous fill into bins with scale_fill_stepsn
Task: Stakeholders sometimes prefer a binned heatmap so categories are clearly distinguishable rather than a smooth gradient. Take the churn heatmap from exercise 1.4 and apply scale_fill_stepsn(colours = c("#fee0d2","#fc9272","#de2d26"), breaks = c(10, 20, 30)) to create four risk bands. Save the plot to ex_2_4.
Expected result:
#> Same 4x3 heatmap as ex_1_4, but fill is binned into 4 bands:
#> <10%, 10-20%, 20-30%, >30%. Color jumps at each break.Difficulty: Advanced
Turn the smooth gradient into a small set of discrete bands so each cell snaps to one of a few colors at named thresholds.
Add scale_fill_stepsn() with a colours vector of band colors and breaks = c(10, 20, 30).
Click to reveal solution
Explanation: scale_fill_stepsn() is the binned cousin of scale_fill_gradientn(). You supply N+1 colors and N breaks; ggplot2 buckets values and assigns each tile to a band color. This is the right scale when business consumers ask "show me the red ones" and you want hard cutoffs at named thresholds (10%, 20%, 30%). The smooth scale_fill_gradient2() blurs the boundary; stepsn makes it auditable. For exploratory work keep the gradient; for executive dashboards bin it.
Section 3. Ordering rows and columns (3 problems)
Exercise 3.1: Reorder y-axis by row total so the busiest day floats to the top
Task: Take the call volume heatmap data calls from exercise 1.1 and reorder the day factor so that the busiest day (highest sum of volume) sits at the top of the y-axis. Use reorder(day, volume, sum) inside aes() then plot as geom_tile() and save the ggplot to ex_3_1.
Expected result:
#> A 7x24 heatmap. y-axis ordered by total daily volume (ascending
#> from bottom). Most-loaded day sits at the top, lightest at the bottom.Difficulty: Intermediate
Rebuild the day axis so its order follows each day's total volume rather than the alphabet.
Inside aes(), set the y mapping to reorder(day, volume, sum).
Click to reveal solution
Explanation: reorder(factor, value, fun) rebuilds the factor levels using fun (here sum) applied to value within each level. The default sort is ascending, so the largest sum lands at the highest position, which on a ggplot2 y-axis means the top. To flip the order use reorder(day, -volume, sum) or wrap with fct_rev(). This pattern is a one-line replacement for first computing totals, then setting levels manually with factor(levels = ...).
Exercise 3.2: Order rows by hierarchical clustering so similar profiles cluster together
Task: A geneticist wants gene expression rows ordered by similarity rather than alphabetically so that co-expressed genes sit next to each other on the heatmap. Cluster a 6-gene by 4-sample inline matrix using hclust(dist(mat)), extract the leaf order, and use it to set the gene factor levels. Plot as a heatmap and save the result to ex_3_2.
Expected result:
#> A 6x4 heatmap. y-axis order: c("g3","g1","g5","g2","g6","g4")
#> (the dendrogram leaf order from hclust on Euclidean distance).Difficulty: Advanced
Let similarity decide the row order: cluster the wide matrix, then reuse its dendrogram leaf sequence as the factor order.
Take hclust(dist(expr_mat))$order, index rownames(expr_mat) with it, and pass that as levels to factor(gene, ...).
Click to reveal solution
Explanation: The two-step trick is to cluster the wide matrix first (hclust(dist(...))) and then use the leaf order to refactor the long data. hclust() returns a $order slot with the row indices in dendrogram traversal order. Skipping the wide-to-long pivot is fine because ggplot2 only needs the leaf order, not the dendrogram itself; if you want to render the dendrogram alongside the heatmap, use the ggdendro package or patchwork::wrap_plots(). For column clustering apply the same trick to t(mat).
Exercise 3.3: Reverse the y-axis so the first level sits at the top
Task: By default ggplot2 puts the first factor level at the bottom of the y-axis. Heatmaps usually read top-down (like a table), so you want the first level at the top. Take the airquality heatmap from exercise 1.2 and reverse the y-axis with scale_y_discrete(limits = rev). Save the plot to ex_3_3.
Expected result:
#> Same airquality heatmap; y-axis now goes Month 5 at top
#> down to Month 9 at the bottom (instead of 5 at bottom).Difficulty: Beginner
Heatmaps read top-down like a table, so flip the discrete y-axis that ggplot2 otherwise builds from the bottom up.
Add scale_y_discrete(limits = rev) to the plot, passing the bare function rev.
Click to reveal solution
Explanation: Passing the function rev (not rev()) to limits is the idiomatic ggplot2 way to flip a discrete axis. The scale function is called internally with the current levels and gets to transform them. The alternative scale_y_discrete(limits = c("9","8","7","6","5")) works but is brittle if the levels change. For continuous axes use scale_y_reverse(). The visual effect matters for heatmaps because most readers scan top-down like a spreadsheet.
Section 4. Labels, annotations, and text overlays (3 problems)
Exercise 4.1: Overlay numeric values on each tile with geom_text
Task: A retail buyer reviewing weekly category sales wants to see the exact number inside each heatmap cell, not just the color. Take the churn matrix churn from exercise 1.4 and add geom_text(aes(label = churn_pct)) so each tile shows its percentage value. Save the resulting plot to ex_4_1.
Expected result:
#> Same 4x3 heatmap as ex_1_4, but each tile now has a numeric label
#> centered on it (e.g. "38", "22", ..., "4"). Default black text.Difficulty: Intermediate
Add a second layer on top of the colored tiles that prints each cell's numeric value at its center.
Add geom_text(aes(label = churn_pct)) after the tile layer.
Click to reveal solution
Explanation: geom_text() inherits x and y from the heatmap layer, so it lines up with the tile centers automatically. The text color is the layer color (default black), which fails on dark-fill cells (next exercise solves that). For percentages format the label with sprintf("%d%%", churn_pct) or scales::label_percent(scale = 1). For thousands separators wrap with scales::label_comma(). Keep labels short; if the value is two digits or fewer they fit comfortably even on small tiles.
Exercise 4.2: Conditionally color labels white on dark cells and black on light cells
Task: Black labels disappear against deep red tiles. Take the labeled churn heatmap and conditionally set the label color to "white" when churn_pct > 20 and "black" otherwise so the text is readable on every cell. Use geom_text(aes(label = churn_pct, colour = churn_pct > 20)) plus scale_colour_manual(). Save the plot to ex_4_2.
Expected result:
#> Same labeled heatmap. Cells with churn > 20% (top-left corner)
#> now show white text; cells with churn <= 20% show black text.Difficulty: Intermediate
Make the label color depend on how dark its cell is, so the text stays legible on both light and dark tiles.
In geom_text() map colour = churn_pct > 20, then bind that boolean to literal colors with scale_colour_manual() and hide its legend.
Click to reveal solution
Explanation: Mapping a boolean to the colour aesthetic creates a two-level discrete scale, which you then bind to literal colors via scale_colour_manual(). The guide = "none" argument hides the legend because this is purely a visual fix, not a piece of information. Choose the threshold to roughly bisect the data; setting it at the visual midpoint of the fill scale (often the 50th percentile or the gradient midpoint) gives the cleanest contrast for any color palette.
Exercise 4.3: Outline a specific cell with geom_rect to highlight an anomaly
Task: A site reliability engineer is reviewing the call volume heatmap and wants to draw a thick black box around the single busiest cell to flag it for the incident write-up. Find the row of calls with the maximum volume, then overlay geom_rect() with xmin/xmax = hour +- 0.5 and ymin/ymax = day +- 0.5 mapped to a 4-element constant. Save the plot to ex_4_3.
Expected result:
#> A 7x24 heatmap with one thick black rectangle outlining the
#> single tile whose volume is the dataset maximum. No fill change.Difficulty: Advanced
Draw an unfilled outline over just the single extreme cell, using numeric coordinates because the day axis is discrete.
Add geom_rect() on the peak data with xmin/xmax and ymin/ymax offset by 0.5, fill = NA, and inherit.aes = FALSE.
Click to reveal solution
Explanation: geom_rect() needs numeric x/y coordinates, but the heatmap y-axis is discrete, so you first convert the day to its integer factor index. inherit.aes = FALSE is essential: without it, geom_rect() tries to inherit fill = volume from the parent ggplot and you get a colored box plus a warning. The same recipe works for any annotation (text callouts, leader lines) on top of a categorical-axis heatmap. For a softer highlight swap to fill = "yellow", alpha = 0.3.
Section 5. Correlation and statistical heatmaps (3 problems)
Exercise 5.1: Correlation heatmap of mtcars with diverging fill
Task: A junior analyst wants the classic mtcars correlation heatmap. Compute cor(mtcars), pivot the resulting matrix into long format with pivot_longer(), then plot with geom_tile() and scale_fill_gradient2() centered on zero. Save the ggplot object to ex_5_1 so it shows all 11 variables on both axes.
Expected result:
#> An 11x11 heatmap of pairwise Pearson correlations.
#> Diagonal = 1 (deep red). mpg vs cyl ~ -0.85 (deep blue).
#> wt vs disp ~ 0.89 (deep red). Diverging palette around 0.Difficulty: Intermediate
Once the correlation matrix is reshaped to long form, it plots like any other three-column heatmap with a zero-centered scale.
Add a tile layer and scale_fill_gradient2(midpoint = 0) to the cor_long data.
Click to reveal solution
Explanation: The three-step recipe (correlate, pivot, plot) is the workhorse for any pairwise statistic, not just correlation. Swap cor() for cor() with method = "spearman" for rank correlation, or for any custom function returning a square matrix. The 45-degree x-axis label is conventional for variable-name labels that would otherwise overlap. To hide the redundant upper triangle, filter cor_long to var1 <= var2 lexicographically (exercise 5.3 explores the masked variant).
Exercise 5.2: Heatmap of Titanic survival counts by Class and Sex
Task: The built-in Titanic array stores counts by Class, Sex, Age, and Survived. Marginalize over Age and Survived (keep only survivors) to produce a 4-class by 2-sex grid of survival counts, then plot as a geom_tile() heatmap with viridis fill and value labels. Save the plot to ex_5_2.
Expected result:
#> A 4x2 heatmap. Highest cell: 1st class / Female ~ 141 survivors
#> (bright yellow). Lowest: 3rd / Male ~ 75 (dark purple).
#> Each tile labelled with its integer count.Difficulty: Intermediate
The summarized surv table is already one row per cell, so map the two categories and the count, then label each tile.
Use aes(x = Sex, y = Class, fill = n) with a tile layer, geom_text(aes(label = n)), and scale_fill_viridis_c().
Click to reveal solution
Explanation: Heatmaps of contingency tables are an underused alternative to grouped bar charts because they keep the cross-classified structure visible. The key transformation is as.data.frame() on the array, which gives one row per cell with a Freq column, then collapse over the unwanted dimensions with group_by() and summarise(). White labels read better than black on the viridis palette; if you mix fill scales, return to the conditional-color pattern from exercise 4.2.
Exercise 5.3: Significance-masked correlation heatmap
Task: A biostatistician wants the mtcars correlation heatmap from exercise 5.1 but with non-significant cells (p >= 0.05 from a pairwise test) blanked out so the eye is drawn only to the reliable correlations. Compute pairwise p-values with cor.test(), set corr to NA where p >= 0.05, and use na.value = "grey90" in the fill scale. Save the plot to ex_5_3.
Expected result:
#> An 11x11 heatmap. ~70% of off-diagonal cells colored red or blue
#> (significant correlations); the remaining cells are light grey
#> (masked because p >= 0.05). Diagonal stays solid red.Difficulty: Advanced
Plot a parallel value column that is blank wherever the correlation is not significant, and let the scale color those blanks with a neutral tone.
Map fill = corr_sig in the tile layer and set na.value = "grey90" inside scale_fill_gradient2().
Click to reveal solution
Explanation: The rowwise() + cor.test() combo computes a p-value per cell without writing an explicit loop. The masking trick is to keep a parallel column (corr_sig) that is NA for non-significant cells; the fill scale's na.value parameter colors those cells with a neutral tone that recedes. A more rigorous approach corrects for multiple testing (Bonferroni or BH) before applying the threshold; with 121 cells and alpha 0.05 you would expect roughly 6 false positives uncorrected.
Section 6. Facets, missing values, and polish (3 problems)
Exercise 6.1: Faceted cohort retention heatmap by acquisition year
Task: A SaaS analyst wants a separate cohort retention heatmap for each cohort_year so leadership can compare the 2022 cohort to the 2023 cohort side by side. Take the inline retention tibble, plot a geom_tile() heatmap of month_number vs cohort_month, and add facet_wrap(~ cohort_year). Save the plot to ex_6_1.
Expected result:
#> Two heatmap panels side by side (cohort_year = 2022 and 2023).
#> Each panel: x = month_number (1..6), y = cohort_month (Jan..Jun),
#> fill = retention_pct. Default blue gradient, brighter early months.Difficulty: Intermediate
Split one heatmap into a small panel per cohort year so the years sit side by side under a shared color scale.
Add facet_wrap(~ cohort_year) to a tile plot of month_number versus cohort_month.
Click to reveal solution
Explanation: facet_wrap() is the canonical way to slice a heatmap across a third (categorical) dimension. Each panel reuses the same fill scale by default so the colors are comparable across cohorts, which is exactly what you want for a "is 2023 better than 2022" read. If the cohorts had wildly different retention ranges, facet_wrap(scales = "free") would let each panel have its own legend, but that almost always confuses the comparison. For a single dense grid prefer one heatmap with the third dimension on one of the axes.
Exercise 6.2: Render missing values with na.value instead of dropping them
Task: Sometimes you want NA cells to remain visible on the heatmap as explicit "no data" rather than being silently dropped. Take the full airquality dataset (do not filter out NA Ozone rows), plot the heatmap, and set na.value = "grey80" inside scale_fill_viridis_c() so NA cells render as light grey. Save the plot to ex_6_2.
Expected result:
#> Month-by-day Ozone heatmap with all 153 cells drawn.
#> ~37 NA cells appear as light grey; non-NA cells use viridis colors.Difficulty: Intermediate
Keep the missing cells on the canvas as explicit "no data" and give them a quiet color instead of dropping them.
Set na.value = "grey80" inside scale_fill_viridis_c().
Click to reveal solution
Explanation: Every continuous fill scale in ggplot2 accepts na.value. The default is "grey50" which is dark enough to fight with most palettes; bumping it to "grey80" or "white" makes missing cells recede. Showing NA cells explicitly is the right choice when missingness is informative (sensor downtime, recording gaps, drop-out) rather than incidental. If you instead want to highlight missing cells, set na.value = "red" and they will jump out for QA review.
Exercise 6.3: Production-quality polish with square tiles, rotated labels, and minimal theme
Task: Take the mtcars correlation heatmap from exercise 5.1 and polish it for a stakeholder report: square tiles via coord_fixed(), x-axis labels rotated 45 degrees with right-justified anchors, remove the panel grid and background using theme_minimal() plus targeted theme() calls, and add a title and a fill legend title. Save the plot to ex_6_3.
Expected result:
#> The 11x11 correlation heatmap, polished: square cells,
#> 45-degree rotated x labels, no panel grid, no axis ticks,
#> title "mtcars correlation matrix", legend titled "Pearson r".Difficulty: Advanced
Square the cells, strip the background clutter, and add report-ready titles and rotated axis labels.
Chain coord_fixed(), theme_minimal(), a theme() that uses element_blank() for the grid and ticks and rotates axis.text.x, plus labs().
Click to reveal solution
Explanation: coord_fixed() locks the aspect ratio so the tiles are perfect squares regardless of plot device size; this matters for correlation matrices where rectangular cells imply a relationship between row and column extent that does not exist. Stripping the panel grid with panel.grid = element_blank() is necessary because theme_minimal() keeps faint gridlines that show through the tiles. The order of theme_minimal() then theme() matters: the later call wins, so put global theme first and the targeted overrides afterwards.
What to do next
- Learn more visualization patterns in the Top 50 ggplot2 Visualizations gallery.
- Practice dplyr verbs that feed these heatmaps in dplyr Exercises in R.
- Reshape skills (pivot_longer, pivot_wider) are drilled in tidyr Exercises in R.
- For categorical color and factor reordering, see forcats Exercises in R.
r-statistics.co · Verifiable credential · Public URL
This document certifies mastery of
ggplot2 Heatmap 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.
126 learners have earned this certificate