import tensorflow as tf from sklearn.metrics import roc_auc_score def auroc(y_true, y_pred): return tf.py_func(roc_auc_score, (y_true, y_pred), tf.double) # Build Model... model.compile(loss='categorical_crossentropy', optimizer='adam',metrics=['accuracy', auroc])