[python] SHAP summary plot,Gray scale, 흑백으로 그리기

fig1. title

1. shap version check and building a explainer

2. matplotlib colormap, make color dictionary

3. shap.summary_plot

---

1. shap version check and building a explainer

 사용한 shap version : 0.37.0

import shap

shap.initjs()
print(shap.__version__)

 building a explainer

# changing names of features
names = pd.read_csv('~/name_dictionary.csv', index_col='var')

background = shap.maskers.Independent(X_train_features_new, max_samples=100)

# type of model is XGBoost, output of explainer's output is probability
explainer = shap.TreeExplainer(ML_model, model_output='probability', data=X_train_features_new)

# calculate the shap values
shap_values = explainer.shap_values(X_train_features_new)

# set the dictionary of feature's name
meaning_dict = names.to_dict()['name']

 

 보통 shap summary plot을 그리면 아래 그림과 같음

# SHAP, train set
shap.summary_plot(shap_values, X_train_features_new.rename(columns=meaning_dict))

fig2. default style of shap.summary plot

---

2. matplotlib colormap, make color dictionary

 기본 스타일도 좋지만 흑백으로 하고 싶을 때 matplotlib.colors를 이용해서 맵핑 작업을 해주면 됨.

#shap gray version configuration
import matplotlib.colors as colors

gray_val=0.75
cdict = {'red': [(0.0, gray_val, gray_val),
                (1.0, 0.0, 0.0)],
        'green': [(0.0, gray_val, gray_val),
                 (1.0, 0.0, 0.0)],
        'blue': [(0.0, gray_val, gray_val),
                (1.0, 0.0, 0.0)]}
gray2black = colors.LinearSegmentedColormap(name='red', segmentdata=cdict)

---

3. shap.summary_plot

# SHAP,test all / gray-scale version

shap_values_all = explainer.shap_values(X_test_features)

shap.summary_plot(shap_values_all, X_test_features.rename(columns=meaning_dict), show=False, 
                  cmap=plt.get_cmap(gray2black))
plt.tight_layout()
plt.show()

fig3. gray-scale style of shap.summary plot