Keras: how to get top-k accuracy

Alternatively,

from keras.metrics import top_k_categorical_accuracy

def topKacc(Y_true, Y_pred):
  return top_k_categorical_accuracy(Y_true, 
                                    Y_pred, 
                                    k = int_here)
metrics = [topKacc, ...]

model.compile(...,
              metrics=metrics)

Ok here is the code that works for me, in case someone else stumbles upon similar issues - the missing link for me was using ".evaluate":

import functools
top3_acc = functools.partial(keras.metrics.top_k_categorical_accuracy, k=3)

top3_acc.__name__ = 'top3_acc'

model.compile(Adam(lr=.001),#
    optimizers.RMSprop(lr=2e-5),
        loss='categorical_crossentropy',
        metrics=['accuracy','top_k_categorical_accuracy',top3_acc])

    model.evaluate(X_test, y_test)

where 'top_k_categorical_accuracy' gives me the score for k=5 (standard) and top3_acc can be adjusted by changing k=3 in the function call.