A waterfall-style chart showing how each feature contributes to pushing a model prediction from a base value (expected model output) to the final predicted value. Horizontal bar segments extend right for positive SHAP values and left for negative SHAP values, stacking cumulatively so the viewer can trace the path from baseline to prediction. This is a core ML explainability visualization for explaining individual predictions, complementing the SHAP summary plot which shows feature effects across many samples.
""" anyplot.ai
shap-waterfall: SHAP Waterfall Plot for Feature Attribution
Library: seaborn 0.13.2 | Python 3.13.13
Quality: 84/100 | Created: 2026-05-07
"""
import os
import sys
# Prevent self-import: remove this script's directory from sys.path
_script_dir = os.path.dirname(os.path.abspath(__file__))
if _script_dir in sys.path:
sys.path.remove(_script_dir)
import matplotlib.patches as mpatches
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import seaborn as sns
# Theme tokens
THEME = os.getenv("ANYPLOT_THEME", "light")
PAGE_BG = "#FAF8F1" if THEME == "light" else "#1A1A17"
ELEVATED_BG = "#FFFDF6" if THEME == "light" else "#242420"
INK = "#1A1A17" if THEME == "light" else "#F0EFE8"
INK_SOFT = "#4A4A44" if THEME == "light" else "#B8B7B0"
INK_MUTED = "#6B6A63" if THEME == "light" else "#A8A79F"
COLOR_POS = "#AE3030" # imprint red — positive SHAP (raises prediction)
COLOR_NEG = "#4467A3" # imprint blue — negative SHAP (lowers prediction)
sns.set_theme(
style="ticks",
rc={
"figure.facecolor": PAGE_BG,
"axes.facecolor": PAGE_BG,
"axes.edgecolor": INK_SOFT,
"axes.labelcolor": INK,
"text.color": INK,
"xtick.color": INK_SOFT,
"ytick.color": INK_SOFT,
"grid.color": INK,
"grid.alpha": 0.10,
"legend.facecolor": ELEVATED_BG,
"legend.edgecolor": INK_SOFT,
},
)
# Data — credit scoring model predicting loan default probability (single applicant)
base_value = 0.32
feature_names = [
"Late Payment Count",
"Credit Score",
"Debt-to-Income Ratio",
"Loan Amount",
"Monthly Income",
"Employment Years",
"Public Records",
"Savings Balance",
"Account Age",
"Credit Lines",
]
# SHAP values sorted by absolute magnitude descending (index 0 = largest impact)
shap_values = np.array([0.24, -0.18, 0.16, 0.11, -0.10, -0.07, 0.06, -0.05, -0.04, 0.03])
final_value = base_value + shap_values.sum()
n = len(feature_names)
# Display: largest |SHAP| at top (y = n-1), smallest at bottom (y = 0)
# Cumulative flow builds bottom-to-top — smallest feature processed first from base_value
shap_btt = shap_values[::-1]
running = np.concatenate([[base_value], base_value + np.cumsum(shap_btt)])
starts_btt = running[:-1]
ends_btt = running[1:]
features_btt = feature_names[::-1]
bar_lefts = np.where(shap_btt >= 0, starts_btt, ends_btt)
bar_widths = np.abs(shap_btt)
bar_colors = [COLOR_POS if v >= 0 else COLOR_NEG for v in shap_btt]
y_pos = np.arange(n)
# Plot
fig, ax = plt.subplots(figsize=(16, 9), facecolor=PAGE_BG)
ax.set_facecolor(PAGE_BG)
# Waterfall bars — dominant feature (top bar, y=n-1) gets thicker edge for focal emphasis
for i, (yi, left, width, color) in enumerate(zip(y_pos, bar_lefts, bar_widths, bar_colors, strict=False)):
is_dominant = i == n - 1
ax.barh(
yi,
width,
left=left,
color=color,
height=0.55,
zorder=3,
edgecolor=INK_SOFT if is_dominant else PAGE_BG,
linewidth=1.5 if is_dominant else 0.5,
)
# Vertical dotted connector lines at cumulative junctions between adjacent bars
for j in range(n - 1):
cx = ends_btt[j]
ax.plot([cx, cx], [j + 0.30, j + 0.70], color=INK_SOFT, linewidth=1.0, linestyle=":", zorder=2, alpha=0.65)
# Connector dots using seaborn scatterplot — mark the running cumulative total at each junction
connector_df = pd.DataFrame({"x": ends_btt[:-1], "y": np.arange(n - 1) + 0.5})
sns.scatterplot(data=connector_df, x="x", y="y", ax=ax, color=INK_SOFT, s=50, zorder=5, alpha=0.75)
# SHAP value labels beside bars
for yi, left, width, shap in zip(y_pos, bar_lefts, bar_widths, shap_btt, strict=False):
label = f"+{shap:.2f}" if shap >= 0 else f"{shap:.2f}"
if shap >= 0:
ax.text(
left + width,
yi,
f" {label}",
ha="left",
va="center",
color=COLOR_POS,
fontsize=14,
fontweight="bold",
zorder=4,
)
else:
ax.text(
left, yi, f"{label} ", ha="right", va="center", color=COLOR_NEG, fontsize=14, fontweight="bold", zorder=4
)
# Reference lines
ax.axvline(base_value, color=INK_SOFT, linewidth=1.5, linestyle="--", zorder=1, alpha=0.85)
ax.axvline(final_value, color=INK, linewidth=2.0, linestyle="-", zorder=1, alpha=0.90)
# Annotations for reference lines — 15pt for adequate secondary text legibility
ax.text(base_value, -0.75, f"E[f(x)] = {base_value:.2f}", ha="center", va="center", fontsize=15, color=INK_SOFT)
ax.text(
final_value,
n - 0.5,
f"f(x) = {final_value:.2f}",
ha="center",
va="center",
fontsize=15,
color=INK,
fontweight="bold",
)
# Axes — dominant feature label is bold to create visual focal point
ax.set_yticks(y_pos)
ax.set_yticklabels(features_btt, fontsize=16)
for label_obj in ax.get_yticklabels():
if label_obj.get_text() == "Late Payment Count":
label_obj.set_fontweight("bold")
ax.tick_params(axis="y", length=0)
ax.tick_params(axis="x", labelsize=16, colors=INK_SOFT)
ax.set_xlabel("Model Output (Default Probability)", fontsize=20, color=INK)
# Subtle x-axis grid
ax.xaxis.grid(True, alpha=0.10, linewidth=0.8, color=INK, zorder=0)
ax.set_axisbelow(True)
# Spine removal using seaborn's despine — idiomatic seaborn API
sns.despine(ax=ax, left=True, top=True, right=True)
ax.spines["bottom"].set_color(INK_SOFT)
# Legend
pos_patch = mpatches.Patch(color=COLOR_POS, label="Positive SHAP (↑ prediction)")
neg_patch = mpatches.Patch(color=COLOR_NEG, label="Negative SHAP (↓ prediction)")
ax.legend(
handles=[pos_patch, neg_patch],
loc="lower right",
fontsize=16,
framealpha=0.9,
facecolor=ELEVATED_BG,
edgecolor=INK_SOFT,
)
# Axis limits
x_all = np.concatenate([starts_btt, ends_btt])
ax.set_xlim(x_all.min() - 0.06, x_all.max() + 0.16)
ax.set_ylim(-1.0, n)
ax.set_title(
"Credit Default Prediction · shap-waterfall · seaborn · anyplot.ai",
fontsize=24,
fontweight="medium",
color=INK,
pad=16,
)
plt.tight_layout()
plt.savefig(f"plot-{THEME}.png", dpi=300, bbox_inches="tight", facecolor=PAGE_BG)
Part of SHAP Waterfall Plot for Feature Attribution on anyplot.ai.