Issue Description Python process was terminated when calling TreeExplainer.shap_values(). The error message is "malloc(): unsorted double linked list corrupted". Then I did more testing. The behavior is not stable. Sometimes the API can be successfully finished. Any idea about the root cause. Is the...

Issue Description I have a 3-layer NN (implemented using torch) for a binary classification problem. I standardise the features using sklearn standard scaler. I wish to explain model using SHAP DeepExplainer and force_plots. The prediction made using model is 0.97 (positive class). The SHAP plot giv...

Problem Description Some code like https://github.com/shap/shap/blob/master/shap/explainers/_kernel.py#L644-L647 uses the logic A_inv = np.linalg.inv(A) x = A_inv @ b instead of the much more stable and (usually) faster x = np.linalg.solve(A, b) Alternative Solutions No response Additional Context O...

Issue Description When I use conda env run the test, I have installed matplotlib, but the shap project throw the error below: ImportError: matplotlib is not installed so plotting is not available! Run pip install matplotlib to fix this. Minimal Reproducible Example import shap shap.initjs() Traceba...

Issue Description It has come to our notice that calculating shap values from the shap package with feature_pertubation = "interventional" is failing additivity check for extremely large depth decision tree / xgboost models. Our experiment - Created a synthetic data following normal distribution w...

Problem Description Hello, i've been using shap plots for a while and i've always had the same problem. Most plots have cmap parameter for changing the colors, but some of them , like shap.plots.waterfall and shap.plots.bar doesn't have support and i've really tried to edit things, and to some ext...

Problem Description Many users are confused by the fact that shapley values sometimes are not adding up to the model output: #2875 #2204 #2785 #2124 #2765 (Partially) this is due to problems like explained in #2124 (comment), #2204 (comment) or #2204 (comment). Fixing all of that probably is not t...

Issue Description The is a bug in shapley values for NGBoost when col_sample < 1. It seems that shapley values excludes some features when col_sample <1 though the the excluded feature is included in trees and has features importance. model.feature_importances_ array([[7.51157818e-01, 2.93312...

Issue Description After preparing the environment as described in CONTRIBUTING.md, some tests fail: tests/explainers/test_deep.py sF.s..... tests/explainers/test_gradient.py F.... Reason: E tensorflow.python.framework.errors_impl.UnimplementedError: 2...

Problem Description The tests are failing with XGBoost 2.0, as per #3254 . A short-term fix could be to just pin the previous version. This issue however is to get our test suite passing against the latest version, which I think would be a preferable long-term solution: https://github.com/dmlc/xgboo...