Plotting ALE, PD, and SHAP on the same plot — Scikit-Explain latest documentation (original) (raw)

Scikit-Explain

import sys, os current_dir = os.getcwd() path = os.path.dirname(current_dir) sys.path.append(path)

Warning : Running this notebook requires installing the bayeshist package (pip install bayeshist)

import skexplain import plotting_config import numpy as np import shap from bayeshist import bayesian_histogram, plot_bayesian_histogram

estimators = skexplain.load_models() X,y = skexplain.load_data() X = X.astype({'urban': 'category', 'rural':'category'})

explainer = skexplain.ExplainToolkit(estimators[0],X=X, y=y,)

ale_1d_ds = explainer.ale(features=['temp2m', 'dwpt2m', 'swave_flux', 'sfc_temp'], n_bootstrap=10, subsample=0.1, n_jobs=4, n_bins=10)

pd_1d_ds = explainer.pd(features=['temp2m', 'dwpt2m', 'swave_flux', 'sfc_temp'], n_bootstrap=10, subsample=0.1, n_jobs=4, n_bins=10)

results = explainer.load('shap_values.nc')

from skexplain.plot.base_plotting import PlotStructure sns.set_theme() sns.set_style('darkgrid', {'axes.grid' : False})

features = ['sfc_temp', 'temp2m', 'swave_flux', 'dwpt2m']

n_panels=len(features) plotter= PlotStructure(BASE_FONT_SIZE = 16) fig, axes = plotter.create_subplots(n_panels=len(features), n_columns=2, figsize=(8,8), dpi=300, wspace=0.4, hspace=0.35)

for ax in axes.flat: ax.grid()

colors = list(sns.color_palette("Set2"))

for ax, feature in zip(axes.flat, features): explainer.plot_ale(ale_1d_ds, features = [feature], estimator_names = ['Random Forest'], add_hist=False, display_feature_names=plotting_config.display_feature_names, display_units = plotting_config.display_units, ax=ax, line_kws = {'line_colors' : [colors[1]], 'linewidth': 2.0}, to_probability=False, )

# Set add_hist == False since it is already included above.
explainer.plot_pd(pd_1d_ds,
              features = [feature],
              estimator_names = ['Random Forest'],
              ax=ax,
              add_hist=False,
              display_feature_names=plotting_config.display_feature_names,
              display_units = plotting_config.display_units,
              line_kws = {'line_colors' : [colors[2]], 'linewidth': 2.0},
              to_probability=False,
             )

#histdata = X.copy()
#histdata['target'] = y

explainer.plot_shap(features=[feature],
                plot_type = 'dependence',
                shap_values=results,
                estimator_name = 'Random Forest',
                display_feature_names=plotting_config.display_feature_names,
                display_units = plotting_config.display_units,
                interaction_index=None,
                color =colors[2],
                s = 3,
                ax=ax,
                alpha=0.5,
                to_probability=False,
)

handles, labels = ax.get_legend_handles_labels()

# plot it up
data = X[feature].values.astype(float)
bin_edges, beta_dist = bayesian_histogram(data, y,)

ax_twin = plotter.make_twin_ax(ax)
plot_bayesian_histogram(bin_edges, beta_dist, ax=ax_twin, color = 'xkcd:rouge', label='Event Rate')

h, l = ax_twin.get_legend_handles_labels()
ylabels = ax_twin.get_yticklabels()
[label.set_color('xkcd:rouge') for label in ylabels]

major_ax = plotter.set_major_axis_labels(fig, xlabel=None, ylabel_left='Feature Effect', ylabel_right='Event Rate [P(y=1|X)]', title=None, ylabel_right_color = 'xkcd:rouge' )

plotter.set_legend(n_panels, fig, ax, major_ax, handles=handles+h, labels=['ALE', 'PD', 'SHAP', 'Event Rate'], bbox_to_anchor = (0.5, -0.3), ncol=4, fontsize=12, ) plotter.add_alphabet_label(n_panels, axes, pos=(0.9, 0.9))