A specialized heatmap visualization for evaluating classification model performance, displaying the counts or proportions of predicted vs actual class labels. The confusion matrix reveals true positives, false positives, true negatives, and false negatives at a glance, making it essential for understanding model behavior, identifying class imbalances, and diagnosing specific misclassification patterns.
""" anyplot.ai
confusion-matrix: Confusion Matrix Heatmap
Library: plotnine 0.15.4 | Python 3.13.13
Quality: 90/100 | Updated: 2026-05-09
"""
import os
import numpy as np
import pandas as pd
from plotnine import (
aes,
coord_fixed,
element_blank,
element_rect,
element_text,
geom_text,
geom_tile,
ggplot,
labs,
scale_color_identity,
scale_fill_gradient,
theme,
theme_minimal,
)
THEME = os.getenv("ANYPLOT_THEME", "light")
PAGE_BG = "#FAF8F1" if THEME == "light" else "#1A1A17"
INK = "#1A1A17" if THEME == "light" else "#F0EFE8"
INK_SOFT = "#4A4A44" if THEME == "light" else "#B8B7B0"
# Data - realistic multi-class classification results
np.random.seed(42)
# Class names for a sentiment analysis classifier
class_names = ["Negative", "Neutral", "Positive"]
n_classes = len(class_names)
# Create a realistic confusion matrix with:
# - Good diagonal (correct predictions)
# - Some confusion between adjacent sentiment classes
confusion_data = np.array(
[
[85, 12, 3], # Negative: mostly correct, some confused with Neutral
[8, 72, 20], # Neutral: harder to classify, confused with both
[4, 15, 81], # Positive: mostly correct, some confused with Neutral
]
)
# Convert to long format for plotnine
rows = []
for i, true_class in enumerate(class_names):
for j, pred_class in enumerate(class_names):
rows.append({"True Label": true_class, "Predicted Label": pred_class, "Count": confusion_data[i, j]})
df = pd.DataFrame(rows)
# Set categorical order (reverse for y-axis to have first class at top)
df["True Label"] = pd.Categorical(df["True Label"], categories=class_names[::-1], ordered=True)
df["Predicted Label"] = pd.Categorical(df["Predicted Label"], categories=class_names, ordered=True)
# Add text color based on count (theme-adaptive: light text on dark cells, dark text on light cells)
threshold = (confusion_data.max() + confusion_data.min()) / 2
light_text = "#F0EFE8" if THEME == "light" else "#FAF8F1"
dark_text = "#1A1A17" if THEME == "light" else "#E8E8E0"
df["text_color"] = df["Count"].apply(lambda x: light_text if x >= threshold else dark_text)
# Create the confusion matrix heatmap
plot = (
ggplot(df, aes(x="Predicted Label", y="True Label", fill="Count"))
+ geom_tile(color=INK_SOFT, size=2)
+ geom_text(aes(label="Count", color="text_color"), size=20, fontweight="bold", show_legend=False)
+ scale_fill_gradient(
low="#e8f4f8" if THEME == "light" else "#2a3f4f",
high="#08519c" if THEME == "light" else "#2ABCCD",
name="Sample Count",
)
+ scale_color_identity()
+ labs(title="confusion-matrix · plotnine · anyplot.ai", x="Predicted Label", y="True Label")
+ coord_fixed(ratio=1)
+ theme_minimal()
+ theme(
figure_size=(12, 12),
plot_background=element_rect(fill=PAGE_BG, color=PAGE_BG),
panel_background=element_rect(fill=PAGE_BG),
text=element_text(size=14, color=INK),
axis_title=element_text(size=22, weight="bold", color=INK),
axis_text=element_text(size=18, color=INK_SOFT),
plot_title=element_text(size=24, weight="bold", ha="center", color=INK),
legend_title=element_text(size=18, color=INK),
legend_text=element_text(size=16, color=INK_SOFT),
legend_background=element_rect(fill=PAGE_BG if THEME == "light" else "#242420", color=INK_SOFT),
panel_grid_major=element_blank(),
panel_grid_minor=element_blank(),
)
)
# Save the plot
plot.save(f"plot-{THEME}.png", dpi=300, verbose=False)
Part of Confusion Matrix Heatmap on anyplot.ai.