如何在Seaborn FacetGrid绘制热力图时访问行列信息并添加填充图案

To add conditional hatch patterns to each heatmap in your FacetGrid—based on the facet's row/column values (a and b) and cell values—you need to access the current facet's metadata inside your heatmap drawing function. Here's how to do it:

Key Steps Explained #

1. Access the Current Facet's Axes: When using map_dataframe, Seaborn automatically passes the current axis (ax) to your drawing function as a keyword argument.
2. Retrieve Facet's a/b Values: Use the FacetGrid's axes_dict (a map of (row_value, col_value) to axis objects) to find which (a, b) pair corresponds to the current axis.
3. Compute Hatch Mask: Apply your custom function (myFunc) to each cell in the heatmap data, using the current a/b values and cell-specific data.
4. Overlay Hatch Pattern: Use ax.pcolor to draw the hatch pattern on top of the heatmap, using a masked array to target only the cells that need hatching.

Full Working Code #

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

# Custom function to determine if a cell gets a hatch pattern
def myFunc(fgA, fgB, d_val, c_val, cell_val):
    # Example logic: Apply hatch if sum of a + b >15 AND cell value >2
    # Adjust this logic to match your specific requirements
    return 5 if (fgA + fgB) > 15 and cell_val > 2 else 0

# Create your sample data
d = np.array(np.meshgrid(np.arange(1,6), np.arange(6,11), np.arange(1,6), np.arange(6,11))).T.reshape(-1,4)
d = np.c_[d, np.random.randint(1,5,625)]
df = pd.DataFrame(d, columns=['a','b','c','d','e'])

# Initialize the FacetGrid
fg = sns.FacetGrid(df, col='b', row='a', margin_titles=True, height=5, aspect=1)

# Define the heatmap drawing function with hatch logic
def draw_heatmap(x_col, y_col, val_col, **kwargs):
    # Get the current axis for this facet
    ax = kwargs.pop('ax')
    # Get the subset of data assigned to this facet
    data = kwargs.pop('data')
    
    # Pivot and sort data to form the heatmap matrix
    heatmap_data = data.pivot(index=y_col, columns=x_col, values=val_col)
    heatmap_data = heatmap_data.sort_index(ascending=False)
    
    # Find the current a and b values for this facet
    for (a_val, b_val), facet_ax in fg.axes_dict.items():
        if facet_ax is ax:
            current_a = a_val
            current_b = b_val
            break
    
    # Calculate which cells need hatching using your custom function
    hatch_mask = heatmap_data.stack().apply(
        lambda cell: myFunc(current_a, current_b, cell.name[0], cell.name[1], cell)
    ).unstack()
    
    # Create a masked array: only values >=5 will show the hatch
    masked_hatch = np.ma.masked_less(hatch_mask.values, 5)[::-1]  # Reverse to align with heatmap's y-axis
    
    # Draw the base heatmap
    sns.heatmap(heatmap_data, ax=ax, **kwargs)
    
    # Overlay the hatch pattern (alpha=0 keeps fill transparent)
    x_coords = np.arange(len(heatmap_data.columns) + 1)
    y_coords = np.arange(len(heatmap_data.index) + 1)
    ax.pcolor(x_coords, y_coords, masked_hatch, hatch='//', alpha=0.)

# Apply the function to all facets in the grid
fg = fg.map_dataframe(draw_heatmap, 'c', 'd', 'e', cbar=False, cmap='viridis', annot=True, fmt=".0f", linewidths=.5)

# Adjust layout and display the plot
plt.tight_layout()
plt.show()

Important Notes #

• Customize myFunc: Modify the logic inside myFunc to match your specific criteria for applying hatch patterns. The function should return a value >=5 when you want a hatch (since we use np.ma.masked_less to mask values below 5).
• Axis Alignment: We reverse the masked_hatch array with [::-1] to match the heatmap's reversed y-axis (where the first row of data appears at the top of the plot).
• Hatch Style: Change the hatch parameter in ax.pcolor to use different patterns (e.g., '\\\\', 'x', '+', '.').

内容的提问来源于stack exchange，提问作者insofern