# From the above grid search using cross validation, it turns out that a tree depth of 5, iterations of 500 and learning rate of 0.1 result in the most optimal RMSE. Hence using that for the final model.
# In[403]:
model = ctb.CatBoostRegressor(depth=5,
iterations=500,
learning_rate=0.1,
verbose=False)
model.fit(X_train, y_train)
# In[404]:
y_pred = model.predict(X_test)
cat_coeff_imp = model.get_feature_importance(type="PredictionValuesChange")
# In[431]:
len(cat_coeff_imp)
top_10_idx_cat = np.argsort(cat_coeff_imp)[-10:]
X_train.columns[top_10_idx_cat]
# In[398]:
explainer = shap.TreeExplainer(model)
shap_values = explainer.shap_values(X_train)
# In[399]:
