def kshap_explainer(lr_classifier, adult_data):
predictor = predict_fcn(predict_type='proba',
clf=lr_classifier,
preproc=adult_data['preprocessor'])
kshap = KernelShap(predictor=predictor,
link='logit',
feature_names=adult_data['metadata']['feature_names'])
kshap.fit(adult_data['X_train'][:100])
return kshap
def explain(model, TrainDF, TestDF, y_col, normalize=True):
X_Train = np.array(TrainDF.drop(columns=[y_col], axis=1).to_numpy())
y_Train = np.array(TrainDF[[y_col]].to_numpy())
X_Test = np.array(TestDF.drop(columns=[y_col], axis=1).to_numpy())
y_Test = np.array(TestDF[[y_col]].to_numpy())
if normalize:
X_Train = preprocessing.scale(X_Train)
X_Test = preprocessing.scale(X_Test)
model.fit(X_Train, y_Train.ravel())
pred_fcn = model.decision_function
np.random.seed(0)
svm_explainer = KernelShap(pred_fcn)
svm_explainer.fit(shap.kmeans(X_Train, 10))
svm_explanation = svm_explainer.explain(X_Test, l1_reg=False)
feature_names = TrainDF.columns.drop([y_col])
classes = set(TrainDF[y_col])
classes_it = iter(classes)
for class_id in range(0, len(classes)):
print(next(classes_it))
plt.figure()
shap.summary_plot(svm_explanation.shap_values[class_id],
X_Test,
feature_names,
show=False)
plt.savefig(
f'shap_summary_plot_binary-{class_id}{"-normalized" if normalize else ""}.png'
)
plt.figure()
shap.summary_plot(svm_explanation.shap_values,
X_Test,
feature_names,
show=False)
plt.savefig(
f'shap_summary_plot_binary_X{"-normalized" if normalize else ""}.png')
def mock_ks_explainer(request):
"""
Instantiates a KernelShap explainer with a mock predictor.
"""
pred_out_dim, link = request.param
predictor = MockPredictor(out_dim=pred_out_dim, seed=0)
explainer = KernelShap(predictor=predictor, seed=0)
return explainer
def mock_kernel_shap_explainer(request):
"""
Instantiates a KernelShap explainer with a mock predictor.
"""
pred_out_dim, link, distributed_opts = request.param
predictor = MockPredictor(out_dim=pred_out_dim, seed=0)
explainer = KernelShap(predictor=predictor,
link=link,
distributed_opts=distributed_opts,
seed=0)
return explainer
def make_kernel_shap(dirname: Optional[Path] = None) -> KernelShap:
np.random.seed(0)
# load data
wine = load_wine()
data = wine.data
target = wine.target
target_names = wine.target_names
feature_names = wine.feature_names
# train classifier
X_train, X_test, y_train, y_test = train_test_split(data,
target,
test_size=0.2,
random_state=0)
scaler = StandardScaler().fit(X_train)
X_train_norm = scaler.transform(X_train)
X_test_norm = scaler.transform(X_test)
classifier = SVC(
kernel="rbf",
C=1,
gamma=0.1,
decision_function_shape="ovr",
# n_cls trained with data from one class as positive
# and remainder of data as neg
random_state=0,
)
classifier.fit(X_train_norm, y_train)
# build kernel shap model
pred_fcn = classifier.decision_function
svm_explainer = KernelShap(pred_fcn)
svm_explainer.fit(X_train_norm)
if dirname is not None:
svm_explainer.save(dirname)
return svm_explainer