Explain Your Machine Learning Model Predictions with GPU-Accelerated SHAP

Technical Blogs & Events Technical Blog
1

Originally published at: https://developer.nvidia.com/blog/explain-your-machine-learning-model-predictions-with-gpu-accelerated-shap/

Learn how to train an XGBoost model, implement the SHAP technique in Python using a CPU and GPU, and compare results between the two.

2

Thanks for checking out my blog on GPU accelerated SHAP values. Let me know if you have any questions about the blog or the code implementation.

3

First off, thanks a lot for such a good blog!
I think I should check something what I missed.

        shap.initjs()
        warnings.filterwarnings(action = "ignore")
        explainer = shap.TreeExplainer(model)
        shap_values = explainer.shap_values(X_test)
        fig = shap.summary_plot(shap_values, X_test, plot_type="bar")
        plt.savefig(os.path.join('shap_image.png'))

At Summary_plot,
it raise

Attempted to set non-positive bottom ylim on a log-scaled axis.
Invalid limit will be ignored.

and the progress never stop. what should I do?

4

Hi @94cogus ,

Firstly, apologies for replying late. I inadvertently missed the notification to your query. As for your issue, I tried it on my end but I encountered no error. Here is the link to the code that I tried let me know if you are still having issues reproducing the results : Google Colab

6

Hi! Can I use a summary plot with the implementation of SHAP on RAPIDS? I already tried, but it doesn’t results. Are there any way? Thank you!