U fb@slddlZddlZddlZddlmZz ddlZWn&ek rDYnedYnXGdddeZ dS)N) Explainerz*xgboost is installed...but failed to load!c@s.eZdZdZdifddZddZddZd S) MimicExplaineraFits a mimic model to the original model and then explains predictions using the mimic model. Tree SHAP allows for very fast SHAP value explainations of flexible gradient boosted decision tree (GBDT) models. Since GBDT models are so flexible we can train them to mimic any black-box model and then using Tree SHAP we can explain them. This won't work well for images, but for any type of problem that GBDTs do reasonable well on, they should also be able to learn how to explain black-box models on the data. This mimic explainer also allows you to use a linear model, but keep in mind that will not do as well at explaining typical non-linear black-box models. In the future we could include other mimic model types given enough demand/help. Finally, we would like to note that this explainer is vaugely inspired by https://arxiv.org/abs/1802.07814 where they learn an explainer that can be applied to any input. Parameters ---------- model : function or iml.Model User supplied function that takes a matrix of samples (# samples x # features) and computes a the output of the model for those samples. The output can be a vector (# samples) or a matrix (# samples x # model outputs). data : numpy.array or pandas.DataFrame or iml.DenseData The background dataset to use for integrating out features. To determine the impact of a feature, that feature is set to "missing" and the change in the model output is observed. Since most models aren't designed to handle arbitrary missing data at test time, we simulate "missing" by replacing the feature with the values it takes in the background dataset. So if the background dataset is a simple sample of all zeros, then we would approximate a feature being missing by setting it to zero. For small problems this background datset can be the whole training set, but for larger problems consider using a single reference value or using the kmeans function to summarize the dataset. xgboostcCs||_||_tt|_t||_tdd|_t ||jd|_ t |j|j |j |j |_ t|j tsptd|j jrtdt|j jdkrtdtt|j jdd d |dS) N keep_indexF)rz;Shap explainer only supports the DenseData input currently.z?Shap explainer does not support transposed DenseData currently.dz Using only z# training data samples could cause zSthe mimic model poorly to fit the real model. Consider using more training samples zPor if you don't have more samples, using shap.inflate(data, N) to generate more.)mimic_model_typemimic_model_paramsZconvert_to_linklinkZconvert_to_modelmodelkwargsgetrZconvert_to_datadataZmatch_model_to_datafZ model_out isinstanceZ DenseDataAssertionErrorZ transposedlenweightslogwarningstr_train_mimic_model)selfr r mimic_modelr rT)Z pred_contribscs"g|]}dd|ddfqS)Nr).0iphirr dsz.MimicExplainer.shap_values..r) rrtypeendswithrrZtreesZpredictrshaperange)rXr rr"r shap_valuesKs  zMimicExplainer.shap_valuesN)__name__ __module__ __qualname____doc__rrr*rrrrrsr) numpynpmultiprocessingsysZ explainerrr ImportErrorprintrrrrrs