Unknown metric error when subclassing tuner

I am trying to subclass a tuner, e.g. Hyperband, to tune batch_size using the following

class MyTuner(kt.tuners.Hyperband):
  def run_trial(self, trial, *args, **kwargs):
    kwargs['batch_size'] = trial.hyperparameters.Int('batch_size', 1024, 4096, step=512)
    super(MyTuner, self).run_trial(trial, *args, **kwargs)

With this, tuner.search gives the following error at the end of the first trial:

Epoch 1/2
59/59 [==============================] - 2s 32ms/step - loss: 5.1755 - accuracy: 0.7180 - val_loss: 0.5713 - val_accuracy: 0.9060
Epoch 2/2
59/59 [==============================] - 2s 29ms/step - loss: 0.4279 - accuracy: 0.9235 - val_loss: 0.2897 - val_accuracy: 0.9425
/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:12: DeprecationWarning: `Tuner.run_trial()` returned None. It should return one of float, dict, keras.callbacks.History, or a list of one of these types. The use case of calling `Tuner.oracle.update_trial()` in `Tuner.run_trial()` is deprecated, and will be removed in the future.
  if sys.path[0] == '':
---------------------------------------------------------------------------

/usr/local/lib/python3.7/dist-packages/keras_tuner/engine/metrics_tracking.py in _assert_exists(self, name)
    284     def _assert_exists(self, name):
    285         if name not in self.metrics:
--> 286             raise ValueError("Unknown metric: %s" % (name,))
    287 
    288 

ValueError: Unknown metric: val_accuracy

The same exact code works if I use Hyperband without subclassing. This happens independently of the model type (tried with a CNN and an LSTM) or the metric used (e.g. tried with accuracy, loss, val_loss).

To Reproduce

https://colab.research.google.com/drive/1-H9PFZwObxpV5Jw04JhlqLHfuz3ncgyJ?usp=sharing

Expected behaviour

Tuning should continue and batch size should be tuned.

[opsxcq]

Found the problem, please change your code to

class MyTuner(kt.tuners.Hyperband):
  def run_trial(self, trial, *args, **kwargs):
    kwargs['batch_size'] = trial.hyperparameters.Int('batch_size', 1024, 4096, step=512)
    return super(MyTuner, self).run_trial(trial, *args, **kwargs)

[BIA4-course]

Just to mention that adding the return statement worked for me.