Seaborn Matrix & Heatmaps

🌡️ heatmap

sns.heatmap(matrix)
sns.heatmap(matrix, annot=True)                      # show values as text in each cell
sns.heatmap(matrix, annot=True, fmt=".2f")               # format annotation numbers
sns.heatmap(matrix, cmap="coolwarm")                         # colormap
sns.heatmap(matrix, cmap="coolwarm", center=0)                  # diverging, centered at 0
sns.heatmap(matrix, vmin=0, vmax=100)                               # fix color scale range
sns.heatmap(matrix, cbar=False)                                        # hide the colorbar
sns.heatmap(matrix, linewidths=0.5, linecolor="white")                    # cell borders

🔗 Correlation matrix (the most common heatmap use case)

corr = df.corr(numeric_only=True)
sns.heatmap(corr, annot=True, cmap="coolwarm", center=0, fmt=".2f")

import numpy as np
mask = np.triu(np.ones_like(corr, dtype=bool))       # mask upper triangle — avoid redundant mirrored info
sns.heatmap(corr, mask=mask, annot=True, cmap="coolwarm", center=0)

Correlation matrices are symmetric — masking the upper triangle removes visual clutter without losing information.

🌳 clustermap — heatmap + hierarchical clustering

sns.clustermap(matrix)
sns.clustermap(matrix, cmap="viridis", standard_scale=1)      # standardize columns before clustering
sns.clustermap(matrix, method="average", metric="euclidean")     # clustering algorithm + distance metric
sns.clustermap(matrix, row_cluster=False)                            # cluster only columns, not rows

clustermap reorders rows/columns to group similar ones together (via dendrograms on the sides) — reveals structure a plain heatmap won't show, since plain heatmap keeps original row/column order.

clustermap is figure-level and manages its own layout entirely — doesn't accept ax=, can't be combined into a subplot grid.

🧮 Pivoting data into matrix shape first

pivot = df.pivot_table(index="day", columns="time", values="total_bill", aggfunc="mean")
sns.heatmap(pivot, annot=True, fmt=".1f")

heatmap expects a 2D matrix (rows × columns of numbers) — long-format DataFrames usually need pivot_table() first.

🔗 Next

Seaborn Multi-Plot Grids · Seaborn Styling & Themes