A SHAP (SHapley Additive exPlanations) summary plot displaying the distribution of SHAP values for each feature, ordered by mean absolute SHAP value (importance). Each dot represents a sample, positioned horizontally by its SHAP value and colored by the feature's value (typically low=blue to high=red). This visualization is essential for machine learning interpretability, showing both feature importance and the direction and magnitude of feature effects on model predictions.
#' anyplot.ai
#' shap-summary: SHAP Summary Plot
#' Library: ggplot2 3.5.1 | R 4.4.1
#' Quality: 86/100 | Created: 2026-09-09
library(ggplot2)
library(dplyr)
library(tibble)
library(scales)
library(ragg)
set.seed(42)
# --- Theme tokens ------------------------------------------------------------
THEME <- Sys.getenv("ANYPLOT_THEME", "light")
PAGE_BG <- if (THEME == "light") "#FAF8F1" else "#1A1A17"
ELEVATED_BG <- if (THEME == "light") "#FFFDF6" else "#242420"
INK <- if (THEME == "light") "#1A1A17" else "#F0EFE8"
INK_SOFT <- if (THEME == "light") "#4A4A44" else "#B8B7B0"
MIDPOINT_BG <- if (THEME == "light") "#FAF8F1" else "#1A1A17"
# Imprint palette-derived diverging cmap (low=blue, mid=neutral, high=matte-red)
# chosen over imprint_seq because low/high feature values are the domain's
# universally recognised polarity (cold->hot / low->high), mirrored here for
# SHAP's own "low feature value -> blue, high -> red" reading convention.
SHAP_LOW <- "#4467A3"
SHAP_HIGH <- "#AE3030"
# --- Data --------------------------------------------------------------------
# Simulated SHAP values from a gradient-boosted house-price model. Each row is
# one (sample, feature) pair: the feature's contribution to that sample's
# predicted price, plus the feature's own value (min-max normalised for color).
n_samples <- 220
make_feature <- function(name, raw, amplitude, noise_sd, shape = c("increasing", "decreasing", "hump")) {
shape <- match.arg(shape)
value_norm <- (raw - min(raw)) / (max(raw) - min(raw))
trend <- switch(shape,
increasing = (value_norm - 0.5) * 2,
decreasing = (0.5 - value_norm) * 2,
hump = (1 - 4 * (value_norm - 0.5)^2 - 0.5) * 2
)
tibble(
feature = name,
shap_value = amplitude * trend + rnorm(n_samples, 0, noise_sd),
feature_value = raw,
feature_value_norm = value_norm
)
}
shap_df <- bind_rows(
make_feature("Living Area (sqft)", rnorm(n_samples, 1900, 650) |> pmax(420),
amplitude = 42000, noise_sd = 7000, shape = "increasing"),
make_feature("Distance to Downtown (km)", runif(n_samples, 1, 25),
amplitude = 16000, noise_sd = 5000, shape = "hump"),
make_feature("School Rating (1-10)", runif(n_samples, 1, 10),
amplitude = 13000, noise_sd = 4500, shape = "increasing"),
make_feature("House Age (years)", rnorm(n_samples, 25, 15) |> pmax(0),
amplitude = 10000, noise_sd = 4000, shape = "decreasing"),
make_feature("Bathrooms", rpois(n_samples, 2) + 1,
amplitude = 7000, noise_sd = 3000, shape = "increasing"),
make_feature("Crime Rate Index", runif(n_samples, 0, 100),
amplitude = 5500, noise_sd = 2500, shape = "decreasing"),
make_feature("Lot Size (sqft)", rnorm(n_samples, 8000, 3000) |> pmax(1500),
amplitude = 4000, noise_sd = 2200, shape = "increasing"),
make_feature("Has Garage", rbinom(n_samples, 1, 0.65),
amplitude = 3000, noise_sd = 1800, shape = "increasing")
)
# Order features by mean |SHAP value| — most important at top
importance <- shap_df |>
group_by(feature) |>
summarise(mean_abs_shap = mean(abs(shap_value)), .groups = "drop")
shap_df <- shap_df |>
left_join(importance, by = "feature") |>
mutate(feature = factor(feature, levels = importance$feature[order(importance$mean_abs_shap)]))
# Call out the top driver feature (highest mean |SHAP value|) with a star
# prefix on its axis label — a light storytelling touch beyond the raw sort.
top_feature <- importance$feature[which.max(importance$mean_abs_shap)]
feature_levels <- levels(shap_df$feature)
axis_labels <- feature_levels
axis_labels[feature_levels == top_feature] <- paste0("★ ", top_feature)
# --- Plot ----------------------------------------------------------------
# shape=21 gives points a stroke independent of fill, so mid-range values
# (fill == PAGE_BG at the diverging midpoint, per style guide) stay visible
# as a faint outlined ring rather than vanishing into the background.
p <- ggplot(shap_df, aes(x = shap_value, y = feature, fill = feature_value_norm)) +
geom_vline(xintercept = 0, color = INK_SOFT, linewidth = 0.5, linetype = "dashed") +
geom_jitter(shape = 21, color = INK_SOFT, stroke = 0.25,
height = 0.30, width = 0, size = 1.9, alpha = 0.55) +
scale_fill_gradient2(
low = SHAP_LOW, mid = MIDPOINT_BG, high = SHAP_HIGH, midpoint = 0.5,
breaks = c(0.06, 0.94), labels = c("Low", "High"),
name = "Feature value",
guide = guide_colorbar(barheight = unit(3.4, "cm"), barwidth = unit(0.35, "cm"))
) +
scale_x_continuous(labels = label_dollar(scale = 1e-3, suffix = "k")) +
scale_y_discrete(labels = axis_labels) +
labs(
x = "SHAP Value (Impact on Predicted Price)",
y = NULL,
title = "shap-summary · r · ggplot2 · anyplot.ai"
) +
theme_minimal(base_size = 8) +
theme(
plot.background = element_rect(fill = PAGE_BG, color = PAGE_BG),
panel.background = element_rect(fill = PAGE_BG, color = NA),
panel.grid.major.y = element_line(color = scales::alpha(INK, 0.18), linewidth = 0.6),
panel.grid.major.x = element_blank(),
panel.grid.minor = element_blank(),
axis.title = element_text(color = INK, size = 10),
axis.text = element_text(color = INK_SOFT, size = 8),
axis.text.y = element_text(color = INK, size = 9),
axis.ticks = element_blank(),
plot.title = element_text(color = INK, size = 12),
legend.background = element_rect(fill = ELEVATED_BG, color = NA),
legend.text = element_text(color = INK_SOFT, size = 8),
legend.title = element_text(color = INK, size = 10),
legend.position = "right",
plot.margin = margin(t = 10, r = 12, b = 8, l = 8)
)
# --- Save --------------------------------------------------------------------
ggsave(
filename = sprintf("plot-%s.png", THEME),
plot = p,
device = ragg::agg_png,
width = 8,
height = 4.5,
units = "in",
dpi = 400
)
Runnable source as JSON, for any HTTP client: https://api.anyplot.ai/specs/shap-summary/ggplot2/code. Any spec id and library id listed in llms-full.txt fit the same URL shape; every URL below is complete and callable.
{
"spec_id": "shap-summary",
"language": "r",
"library": "ggplot2",
"page": "https://anyplot.ai/shap-summary/r/ggplot2",
"hub": "https://anyplot.ai/shap-summary",
"code_json": "https://api.anyplot.ai/specs/shap-summary/ggplot2/code",
"spec_json": "https://api.anyplot.ai/specs/shap-summary",
"render_light_png": "https://storage.googleapis.com/anyplot-images/plots/shap-summary/r/ggplot2/plot-light.png",
"render_dark_png": "https://storage.googleapis.com/anyplot-images/plots/shap-summary/r/ggplot2/plot-dark.png",
"quality_score": 86.0,
"license": "MIT",
"guide": "https://anyplot.ai/llms.txt"
}Part of SHAP Summary Plot on anyplot.ai.