Create a custom scorer in sklearn

sklearn_custom_scorer_example.py

def specificity_scorer(y_true, y_pred):
  tn, fp, fn, tp = metrics.confusion_matrix(y_true, y_pred).ravel()
  return tn / (tn+fp)

specificity = metrics.make_scorer(specificity_scorer, greater_is_better=True)

# use in cross_val_score
cv = cross_val_score(xgb, dt.features_.values, y, scoring=specificity, cv=5, verbose=2)

# use in GridSearchCV
grid = GridSearchCV(clf, {}, scoring=specificity)
grid.fit(X_train, y_train)