I'm pretty puzzled by what's happening here and was hoping someone might shed some light on it. We have an XGB Resgressor model saved as a JSON file. We reload that model. xgb_cv = XGBRegressor() xgb_cv.load_model(model_name) Which seems to work fine: XGBRegressor(base_score=0.5, booster='gbtree',...

It has been a long time I am using SHAP package in Python and had no problem importing it. But today I get error when trying to import SHAP and I cannot figure out how to solve this issue. My shap version is 0.39.0 and my python is 3.8 2022-09-29 14:59:27.468815: W tensorflow/stream_executor/platfor...

I used the following codes to draw a waterfall plot. explainer = shap.TreeExplainer(gbm, data=None) shap_values = explainer.shap_values(P) # visualize the first prediction's explanation shap.waterfall_plot(explainer.expected_value[0], shap_values[0]) It shows this error AttributeError: 'numpy.floa...

Turning on filterwarnings = error in pytest errors when importing shap with ipython >= 8: .tox/py39/lib/python3.9/site-packages/shap/__init__.py:38: in <module> from .plots._beeswarm import summary_legacy as summary_plot .tox/py39/lib/python3.9/site-packages/shap/plots/__init__.py:6: in...

I was using the version 0.40.0 of SHAP with an old version of keras version and it worked like a charm. However, I had to migrate to a new computer and now I am getting the following error: AssertionError: The SHAP explanations do not sum up to the model's output! This is either because of a roundin...

Hello, I would like to know why we need to give both in the TreeExplainer and when calculating the shap values the features as input parameters: explainer = shap.TreeExplainer(model, features, feature_perturbation="interventional",model_output="probability",) shap_values = explainer(features) Previ...

I'm working with an explanation object which I obtained from a model with a One-Hot encoder step. I'm working with the following versions: shap==0.41.0 scikit-learn==1.1.1 To obtain the shap explanation I did the following (it's a MWE, you can reproduce it): import pandas as pd import numpy as np...

As explained here, TreeSHAP uses a conditional background distribution, which might cause some problems when the noisy features are correlated with some signal features. I wonder if exists a variant of TreeSHAP which uses a marginal/unconditional background distribution instead?

Hi All, I'm trying to recreate the input/output heatmap below, but I'm unable to reproduce it. I ran all of the same cells in example provided: https://github.com/slundberg/shap/blob/master/notebooks/text_examples/text_entailment/Textual%20Entailment%20Explanation%20Demo.ipynb The output that gets...

Lets say I have a pandas DataFrame that contains some features and a dependent variable... df_testing_shap =  index y_val feature_1 feature_2 feature_3 1 10.041427 47 0.434511 6 2 9.841583 50 0.452915 9 3 9.713457 34 0.564274 9 Now lets train our model and generate SHAP values (I...