xaicsv 0.1.0

Functions to convert XAI like LIME & SHAP Weights into Dictionary Object for easy CSV manipulation.

About The Project

XaiCSV is a collection of functions to convert XAI (eXplainable AI) like SHAP & LIME into Dictionary Object. The resulting Dictionary is used for easy CSV manipulation such as using Pandas. This package also includes visualization for the Dictionary. Example of Dictionary result :

 {
 "class1": [
     ("feature1",weights1),
     ("feature2",weights2)
 ],
 "class2": [...]
 }

Getting Started

Prerequisites

Python >=3.9 Required

Installation

pip install xaicsv

Usage

1. Use lime_values_to_weights_dict to get Dictionary of LIME Weights.
2. Use shap_values_to_weights_dict to get Dictionary of SHAP Weights.
3. Use weights_dict_to_pyplot to visualize Dictionary values.

Import the xai_csv_utils

 from xaicsv from xai_csv_utils

LIME

1. lime_explanation . LIME Explanation Object.
2. class_names . list of classes . set ["target"] if Regression.

Using iris dataset for classification task and using SVC model.

 list_labels = iris.target_names.tolist()
 list_index_labels= list(range(0, len(list_labels)))

 lime_weights = []
 for i in range(len(df_test)):
 exp = explainer.explain_instance(df_test.iloc[0].values,
                                 svc_iris.predict_proba,
                                 num_features=len(iris.feature_names),
                                 labels= list_index_labels)

 weight_val = xai_csv_utils.lime_values_to_weights_dict(
                                         lime_explanation=exp,
                                         class_names=list_labels)
 lime_weights.append(weight_val)

Add the resulting Dictionary object into a Dataframe

 df['lime_weights'] = lime_weights

SHAP

1. classification . set True if Classification Task. set False if Regression Task.
2. shap_values . SHAP values of one sample.
3. class_names . list of classes . set ["target"] if Regression.
4. feature_names . list of feature names.

Using iris dataset for classification task and using SVC model.

 explainer_shap = shap.KernelExplainer(svc_iris.predict_proba,
                                     train,
                                     feature_names = iris.feature_names)
 shap_values = explainer_shap.shap_values(test)

 list_labels = iris.target_names.tolist()

 shap_weights = []
 for i in range(len(shap_values)):
 weight_val = shap_values_to_weights_dict(class_names=list_labels,
                             classification=True,
                             shap_values=shap_values[i],
                             feature_names=explainer_shap.data_feature_names)

 shap_weights.append(weight_val)

Add the resulting Dictionary object into a Dataframe

 df['shap_weights'] = shap_weights

Weights Visualization

Add the resulting Dictionary object into a Dataframe

 sample_weights = df.iloc[25]["shap_weights"]
 for label in (list_labels):
     fig = weights_dict_to_pyplot(sample_weights, label)