[Day 36] Interpreting ML Model

https://www.codestates.com/

코딩부트캠프 | 코드스테이츠 - 비전공생도 개발자가 될 수 있습니다

---

Warm-up

• PDP ( Partial Dependence Plot )

---

Note

• 부분 의존도 그림 ( Partical Dependence Plots, PDP )

import sklearn
import xgboost
import shap
from sklearn.model_selection import train_test_split

shap.initjs();

df, target = shap.datasets.boston()
X_train,X_test,y_train,y_test = train_test_split(df, target, test_size=0.2, random_state=2)

model = xgboost.XGBRegressor().fit(X_train, y_train)

### Draw PDP plots ###
import matplotlib.pyplot as plt

plt.rcParams['figure.dpi'] = 144

from pdpbox.pdp import pdp_isolate, pdp_plot

feature = 'LSTAT'

isolated = pdp_isolate(
    model=model, 
    dataset=X_test, 
    model_features=X_test.columns, 
    feature=feature,
    grid_type='percentile', # default='percentile', or 'equal'
    num_grid_points=10 # default=10
)
pdp_plot(isolated, feature_name=feature)

• ICE ( Individual Conditional Expectation ) Curves
• SHAP

explainer = shap.TreeExplainer(model)
row = X_test.iloc[[1]]  
shap_values = explainer.shap_values(row)
### Draw SHAP plots ###
shap.initjs()
shap.force_plot(
    base_value=explainer.expected_value, 
    shap_values=shap_values,
    features=row
)

shap_values = explainer.shap_values(X_train)
shap.summary_plot(shap_values, X_train, plot_type='bar')

 

---

Review

오늘은 부분 의존도와 SHAP로 타겟과의 관계를 시각화 해보았다.

---

참고

https://www.youtube.com/watch?v=21QAKe2PDkk