Heatmap not respecting the color bounds

Question

For the following dataframe of a chi2 correlation study, i started to plot a heatmap:

import pandas as pd
import numpy as np


columns = ['A', 'B', 'C', 'D', 'E', 'F', 'G']


results = np.array([[0.70709269, 0.17683162, 0.38328705, 0.61449242, 0.43709035, 0.33675627, 0.2661715 ],
                    [0.17683162, 0.70709268, 0.20520211, 0.16044232, 0.07607822, 0.13364355, 0.13093324],
                    [0.38328705, 0.20520211, 0.81649658, 0.37683897, 0.17308779, 0.29541159, 0.29975079],
                    [0.61449242, 0.16044232, 0.37683897, 0.81649658, 0.4991043 , 0.34257853, 0.2786975 ],
                    [0.43709035, 0.07607822, 0.17308779, 0.4991043 , 0.81649658, 0.22700152, 0.17041603],
                    [0.33675627, 0.13364355, 0.29541159, 0.34257853, 0.22700152, 0.81649658, 0.22018705],
                    [0.2661715 , 0.13093324, 0.29975079, 0.2786975 , 0.17041603, 0.22018705, 0.81649658]])


df_matrix = pd.DataFrame(results, columns=columns)
category_bounds = [0, 0.2, 0.4, 0.6, 0.8, 1.0]
categories = ['Very Weak', 'Weak', 'Moderate', 'Strong', 'Very Strong']


df_heatmap = pd.DataFrame(df_matrix, index=df_matrix.index, columns=df_matrix.columns)


colors = sns.color_palette('coolwarm', len(categories))
cmap = ListedColormap(colors)


fig, ax = plt.subplots(figsize=(10, 8))


sns.heatmap(df_heatmap, annot=True, cmap=cmap, fmt=".3f", cbar=False, ax=ax, linecolor='white')


plt.subplots_adjust(left=0.25, top=0.95)



plt.show()

But, for some reason (I suppose it is due to rounding the values), 0.71 and 0.82 are plotting in the same color. Can someone give me some guidance on what the problem is?

Answer 1

The issue you're facing with the heatmap is likely due to the limited number of colors in the coolwarm colormap. Since you have five categories ('Very Weak', 'Weak', 'Moderate', 'Strong', 'Very Strong'), you need five distinct colors in the colormap to represent each category uniquely.

One way to fix this issue is to use a different colormap that has enough distinct colors to represent all your categories. You can use viridis or plasma colormap, which are perceptually uniform and have enough colors for your categories.

import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt


df_matrix = pd.DataFrame(results, columns=columns)
category_bounds = [0, 0.2, 0.4, 0.6, 0.8, 1.0]
categories = ['Very Weak', 'Weak', 'Moderate', 'Strong', 'Very Strong']

df_heatmap = pd.DataFrame(df_matrix, index=df_matrix.index, columns=df_matrix.columns)

# Using 'viridis' colormap
cmap = 'viridis'

fig, ax = plt.subplots(figsize=(10, 8))

sns.heatmap(df_heatmap, 
            annot=True, 
            cmap=cmap, 
            fmt=".3f", 
            cbar=False, 
            ax=ax, 
            linecolor='white')

plt.subplots_adjust(left=0.25, top=0.95)
plt.show()

Answer 2

Setting the vmin and the vmax will make sure that the bounds of the values are spaced evenly by 0.2 with the 5 categories specified.

cmap = sns.color_palette('mako', len(categories), as_cmap=True)


fig, ax = plt.subplots(figsize=(10, 8))


sns.heatmap(df_heatmap, annot=True, cmap=cmap, fmt=".3f", cbar=False, ax=ax, linecolor='white', vmin=0, vmax=1)

Answer 3

The problem, as @yuckyh said, was that seaborn normalizes the variables to produce the colors. To avoid that, setting vmin and vmax was necessary:

colors = sns.color_palette('coolwarm', len(categories))
cmap = ListedColormap(colors)

# Adjust the figure size to avoid overlapping y-axis labels
fig, ax = plt.subplots(figsize=(10, 8))

# Plot the heatmap
sns.heatmap(df_matrix,
            cmap=cmap,
            fmt=".3f",
            cbar=False,
            ax=ax,
            linecolor='white',
            vmin=0,
            vmax=1,
           annot = True)


# Adjust the position of the lines indicating the edges of the rectangles
ax.hlines(np.arange(n_variables+1), *ax.get_xlim(), color='white', linewidth=1)
ax.vlines(np.arange(n_variables+1), *ax.get_ylim(), color='white', linewidth=1)

ax.set_xticks(np.arange(df_matrix.shape[1]) + 0.5, minor=False)
ax.set_yticks(np.arange(df_matrix.shape[0]) + 0.5, minor=False)

# Configure the tick labels
ax.set_xticklabels(df_heatmap.columns, rotation=45, ha="right")
ax.set_yticklabels(df_heatmap.index, rotation=0)

# Create a custom legend
legend_labels = [f'{category}: {category_bounds[i]:.1f} - {category_bounds[i+1]:.1f}' for i, category in enumerate(categories)]
legend_elements = [plt.Rectangle((0, 0), 1, 1, fc=colors[i]) for i in range(len(categories))]

# Reverse the order of the legend elements and labels
legend_elements = legend_elements[::-1]
legend_labels = legend_labels[::-1]

# Add the legend
ax.legend(handles=legend_elements, labels=legend_labels, loc='center left', bbox_to_anchor=(1, 0.5))

# Adjust the spacing to avoid overlapping y-axis labels
plt.subplots_adjust(left=0.25, top=0.95)

plt.show()