如何使用 TreeExplainer shap 图表的 y 轴标签创建列表？

Question

How to create a list with the y-axis labels of a TreeExplainer shap chart?

Hello,

I was able to generate a chart that sorts my variables by order of importance on the y-axis. It is an impotant solution to visualize in graph form, but now I need to extract the list of ordered variables as they are on the y-axis of the graph. Does anyone know how to do this? I put here an example picture.

Obs.: Sorry, I was not able to add a minimal reproducible example. I don't know how to paste the Jupyter Notebook cells here, so I've pasted below the link to the code shared via Github.

In this example, the list would be "vB0, mB1, vB1, mB2, mB0, vB2".

minimal reproducible example

    ## SHAP GRAPHIC

import pandas as pd
import seaborn as sns
import numpy as np  # for sample data

import matplotlib.pyplot as plt

from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import cross_val_score
from sklearn.inspection import permutation_importance
import shap
from matplotlib import pyplot as plt

# set seed for reproducibility
np.random.seed(1)

# create arrays of random sample data
cl = np.random.choice(range(1, 6), size=(100, 1))
d = np.random.random_sample(size=(100, 6))

# combine the two arrays
data = np.concatenate([cl, d], axis=1)

# create a dataframe
data = pd.DataFrame(data, columns=['classe', 'mB0', 'mB1', 'mB2', 'vB0', 'vB1', 'vB2'])

# create an 'id' column with sequential numbering 
#fonte: https://pythonexamples.org/pandas-set-column-as-index/#:~:text=Pandas%20%E2%80%93%20Set%20Column%20as%20Index&text=To%20set%20a%20column%20as,index%2C%20to%20set_index()%20method.
data['id'] = data.index

#Specifying Predictors (X) and Target Variable (y)
X = data.drop(['classe','id'], axis=1)   #assigning predictors to X
y = data['classe']                #subsetting only the target variable to y



# implementing train-test-split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=1, stratify=y)

X_test.shape, y_test.shape


#Creating the RF Classifier model with 100 trees
rf = RandomForestClassifier(n_estimators=100)
#Fitting the classifier to the data
rf.fit(X_train, y_train)

#Getting predictions
y_rf_pred = rf.predict(X_test)

##--------------------------- Random Forests Classification ---------------------------

#Creating the RF Classifier model with 100 trees
rf = RandomForestClassifier(n_estimators=100)
#Fitting the classifier to the data
rf.fit(X_train, y_train)

#Getting predictions
y_rf_pred = rf.predict(X_test)

## Finding optimal parameters for this model
from sklearn.model_selection import RandomizedSearchCV
# number of trees in random forest
n_estimators = [int(x) for x in np.linspace(start = 100, stop = 500, num = 10)]
# number of features at every split
max_features = ['auto', 'sqrt']

# max depth
max_depth = [int(x) for x in np.linspace(1, 50, num = 11)]
max_depth.append(None)
# create random grid
random_grid = {
 'n_estimators': n_estimators,
 'max_features': max_features,
 'max_depth': max_depth
 }
# Random search of parameters
rfc_random = RandomizedSearchCV(estimator = rf, param_distributions = random_grid, n_iter = 100, cv = 10, verbose=2, random_state=42, n_jobs = -1, return_train_score=True)
# Fit the model
rfc_random.fit(X_train, y_train)

cross_val = rfc_random.cv_results_


### Running the model with optimized hyperparameters
rf_tuned = RandomForestClassifier(n_estimators=144, max_depth=10, max_features='sqrt')
rf_tuned.fit(X_train,y_train)
y_pred_rftuned = rf_tuned.predict(X_test)
rf_cv_score = cross_val_score(rf_tuned, X, y, cv=10)

# #Summarized classification report
# print("=== Classification Report ===")
# from sklearn.metrics import classification_report, confusion_matrix, ConfusionMatrixDisplay
# print(classification_report(y_test, y_pred_rftuned, digits=4))

# print('\n')

# print("=== Confusion Matrix ===")
# print(confusion_matrix(y_test, y_pred_rftuned))

# cm = confusion_matrix(y_test, y_pred_rftuned)
# disp = ConfusionMatrixDisplay(confusion_matrix=cm, display_labels=rf_tuned.classes_)


# ##Saving the tuned model to a external file to be used again without having to train all over again
# from joblib import dump, load
# dump(rf_tuned, 'test3_rf_classifier.joblib')

## Feature Importance Computed with SHAP Values

explainer = shap.TreeExplainer(rf_tuned)
shap_values = explainer.shap_values(X_test, approximate=False, check_additivity=False)

shap.summary_plot(shap_values, X_test, max_display=1000)

#print ordered columns according to shap values
sv = np.array(shap_values)
sv_mean=np.abs(sv).mean(1).sum(0)
order = np.argsort(sv_mean)[::-1]
ordered_cols = X_test.columns[order]
print(ordered_cols)

Answer 1

TL;DR

sv = np.array(shap_values)
sv_mean=np.abs(sv).mean(1).sum(0)
order = np.argsort(sv_mean)[::-1]
ordered_cols = X_test.columns[order]
print(ordered_cols)

Append the above code block to the notebook you linked and you'll get what you want.

Full answer

summary_plot shows columns ordered by sum of mean of abs of SHAP values. You can't extract them from plot, but you can calculate them.

Fully reproducible example:

from sklearn.datasets import make_classification
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from shap import TreeExplainer, summary_plot

X, y = make_classification(n_samples=1000, n_features=30,
          n_classes=5, n_informative=10, random_state=42)
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42)
rf_clf = RandomForestClassifier(random_state=42)
rf_clf.fit(X_train, y_train)

explainer = TreeExplainer(rf_clf)
shap_values = explainer.shap_values(X_test)
summary_plot(shap_values, max_display=10)

sv = np.array(shap_values)
sv_mean=np.abs(sv).mean(1).sum(0)
order = np.argsort(sv_mean)[::-1]
print(order)

---

array([17, 15, 26,  4, 24,  3, 20,  1,  2, 27, 10, 12, 22, 11, 21, 23,  9,
       28, 19,  7, 16,  8,  5, 14, 25,  0, 13,  6, 29, 18])