How to calculate SHAP values for simple MLP Neural Network?

[pplonski]

I'm building AutoML package where I provide SHAP explanations for different models. I have a problem with SHAP explanations for Neural Networks (issue). The NN that I construct is simple Multi-Layer Perceptron with 2 hidden layers. The computation of SHAP values takes a lot of time (several hours) for very small datasets.

Below simple example:

import numpy as np
import pandas as pd
import keras
import shap
from sklearn import datasets

from keras.optimizers import SGD
from keras.models import Sequential
from keras.layers import Dense

print(keras.__version__) 
print(shap.__version__)

n_features = 10


X, y = datasets.make_classification(
            n_samples=1000,
            n_features=n_features,
            n_informative=4,
            n_redundant=1,
            n_classes=2,
            n_clusters_per_class=3,
            n_repeated=0,
            shuffle=False,
            random_state=0,
        )

model = Sequential()
model.add(Dense(10, activation="relu", input_dim=n_features))
model.add(Dense(10, activation="relu", input_dim=n_features))
model.add(Dense(1, activation="sigmoid"))
model.compile(optimizer=SGD(), loss="binary_crossentropy")
model.fit(X, y)

explainer = shap.KernelExplainer(model.predict_proba, X)
shap_values = explainer.shap_values(X)

keras version = 2.3.1
shap version = 0.35.0

On this simple dataset, computing SHAP values take > 8 hours. What is the faster way to compute the SHAP values? For other algorithms (Xgboost, CatBoost, Extra Tress, LightGBM, Random Forest, Linear Regression) all computations are under 1 minute.

[pplonski]

any update on this one

[tkmamidi]

any update, please?

[Daanielvb]

In my case, I'm not even able to run the KernelExplainer on a jupyter notebook. It crashed using all the RAM.

I'm using MLP + TF-IDF on a 3000 samples and ~9000 words vocabulary. What do you suggest?

[github-actions]

This issue has been inactive for two years, so it's been automatically marked as 'stale'.

We value your input! If this issue is still relevant, please leave a comment below. This will remove the 'stale' label and keep it open.

If there's no activity in the next 90 days the issue will be closed.

[github-actions]

This issue has been automatically closed due to lack of recent activity.

Your input is important to us! Please feel free to open a new issue if the problem persists or becomes relevant again.