Last active
November 23, 2022 07:32
-
-
Save eddjberry/3c1818a780d3cb17390744d6e215ba4d to your computer and use it in GitHub Desktop.
Plot a grid of shap.dependence_plots
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
# Dependencies ---------------------- | |
import math | |
import shap | |
import matplotlib.pyplot as plt | |
# shap_dependence_plot_grid --------- | |
def shap_dependence_plot_grid(cols, | |
shap_values, | |
X, | |
interaction_index = None, | |
alpha = 0.75, | |
xmin = 'percentile(1)', | |
xmax = 'percentile(99)'): | |
''' | |
Parameters | |
---------- | |
cols: Either a list/array of column names or indices | |
shap_values: Your shap values | |
X: The feature data | |
interaction_index: Column for interactions or 'auto' | |
(see help(shap.dependence_plot) for details) | |
alpha: Alpha blending for the points | |
xmin: Either None, a value or a string like 'percentile(1)' | |
xmax: Either None, a value or a string like 'percentile(99)' | |
Returns | |
---------- | |
A grid of plots with 3 columns and as many rows as necessary | |
References | |
---------- | |
Partially inspired by | |
https://stackoverflow.com/a/65710424 (CC BY SA 4.0) | |
''' | |
# get the required number of rows | |
nrows = math.ceil(len(cols) / 3) | |
# create the grid of suplots | |
fig, axes = plt.subplots( | |
nrows = nrows, | |
ncols = 3, | |
figsize = (12, 0.75 + 3 * nrows)) | |
# Turn the axes into a 1d array | |
axes = axes.ravel() | |
# loop through the cols and their indices | |
# plotting a shap.dependence_plot | |
for i, col in enumerate(cols): | |
shap.dependence_plot(col, | |
shap_values, | |
X, | |
interaction_index = interaction_index, | |
alpha = alpha, | |
xmin = xmin, | |
xmax = xmax, | |
ax=axes[i], | |
show=False) | |
# adjust the layout | |
fig.tight_layout(pad = 3) | |
return fig |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment