带有交叉验证的形状值 - 我是否在轴0或轴1上实现shap值？

Question

我能问一下吗，我的目的是使用带有交叉验证的外形来确定模型的最重要功能。

我有此代码：

from sklearn.model_selection import train_test_split
from sklearn.datasets import load_breast_cancer
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import KFold
import shap
import pandas as pd
import numpy as np


#loading and preparing the data
iris = load_breast_cancer()
X = iris.data
y = iris.target
columns = iris.feature_names
#if you don't shuffle you wont need to keep track of test_index, but I think 
#it is always good practice to shuffle your data
kf = KFold(n_splits=2,shuffle=True)

list_shap_values = list()
list_test_sets = list()
for train_index, test_index in kf.split(X):
    X_train, X_test = X[train_index], X[test_index]
    y_train, y_test = y[train_index], y[test_index]
    X_train = pd.DataFrame(X_train,columns=columns)
    X_test = pd.DataFrame(X_test,columns=columns)

    #training model
    clf = RandomForestClassifier(random_state=0)
    clf.fit(X_train, y_train)

    #explaining model
    explainer = shap.TreeExplainer(clf)
    shap_values = explainer.shap_values(X_test)
    #for each iteration we save the test_set index and the shap_values
    list_shap_values.append(shap_values)
    list_test_sets.append(test_index)


#combining results from all iterations
test_set = list_test_sets[0]
shap_values = np.array(list_shap_values[0])

for i in range(1,len(list_test_sets)):
    test_set = np.concatenate((test_set,list_test_sets[i]),axis=0)
    shap_values = np.concatenate((shap_values,np.array(list_shap_values[i])),axis=1)

#bringing back variable names    
X_test_df = pd.DataFrame(X[test_set],columns=columns)
cols = X_test_df.columns
sv = np.abs(shap_values[1,:,:]).mean(0)

importance_df = pd.DataFrame({
    "column_name": cols,
    "shap_values": sv
})

#expected result
importance_df.sort_values("shap_values", ascending=False)

print(importance_df)

我可以问，我是否正确实施了？具体来说，这条线正确吗？

    test_set = np.concatenate((test_set,list_test_sets[i]),axis=0)
    shap_values = np.concatenate((shap_values,np.array(list_shap_values[i])),axis=1)

我在示例代码，但我不明白为什么我将轴0用于test_set和轴1用于塑形值。我问了一个问题，我有一个错误在这里，这是在评论中出现的，但是我不清楚地了解如何正确编码这一suggeestion。

Answer 1

我会这样做：

from sklearn.model_selection import train_test_split
from sklearn.datasets import load_breast_cancer
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import KFold
import shap
import pandas as pd
import numpy as np


#loading and preparing the data
iris = load_breast_cancer()
X = iris.data
y = iris.target
columns = iris.feature_names
#if you don't shuffle you wont need to keep track of test_index, but I think 
#it is always good practice to shuffle your data
kf = KFold(n_splits=2,shuffle=True)

list_shap_values = list()
list_test_sets = list()
for train_index, test_index in kf.split(X):
    X_train, X_test = X[train_index], X[test_index]
    y_train, y_test = y[train_index], y[test_index]
    X_train = pd.DataFrame(X_train,columns=columns)
    X_test = pd.DataFrame(X_test,columns=columns)

    #training model
    clf = RandomForestClassifier(random_state=0)
    clf.fit(X_train, y_train)

    #explaining model
    explainer = shap.TreeExplainer(clf)
    shap_values = explainer.shap_values(X_test)
    #for each iteration we save the test_set index and the shap_values
    list_shap_values.append(shap_values)


# flatten list of lists, pick the sv for 1 class, stack the result
shap_values = np.vstack([sv[1] for sv in list_shap_values])
sv = np.abs(shap_values).mean(0)  # <-- error corrected    
importance_df = pd.DataFrame({
    "column_name": columns,
    "shap_values": sv
})