2

I am having difficulties applying any callbacks to Keras Tuner hyperparameter optimsier objects. Here is the code I run:

from keras.callbacks import TensorBoard, EarlyStopping
%load_ext tensorboard

BATCH_SIZE = 32

time_stamp = time.time()
tensorboard = TensorBoard(log_dir = " graphs/{}".format(time_stamp))
checkpoint = ModelCheckpoint(filepath = r"D:\Uni work\...\CNN.hdf5" , monitor = 'val_accuracy', verbose = 1, save_best_only = True )
early_stopping = EarlyStopping( monitor="val_loss" , patience= 3, verbose=2)

tuner = BayesianOptimization(build_model, objective = "val_accuracy", max_trials = 30, num_initial_points=2,  project_name ="audio_classifier")

tuner.search(x = train_X, y=y_cat_encoded, epochs=35, callbacks =  early_stopping, batch_size = BATCH_SIZE, validation_data = (validation_X, y_validation_cat_encoded))

whilst I would like to apply the tensorboard and checkpoint callbacks, it fails simply by passing the early stopping callback. I get the following error:

C:\Anaconda\envs\test\lib\site-packages\kerastuner\engine\tuner.py in _deepcopy_callbacks(self, callbacks)
    277             callbacks = copy.deepcopy(callbacks)
    278         except:
--> 279             raise ValueError(
    280                 'All callbacks used during a search '
    281                 'should be deep-copyable (since they are '

ValueError: All callbacks used during a search should be deep-copyable (since they are reused across trials). It is not possible to do `copy.deepcopy(<tensorflow.python.keras.callbacks.EarlyStopping object at 0x000001802D138100>)

I am not familiar with the term deep-copyable and what it is suggesting in terms of faulty code. Is anyone familiar with how to address this problem?

ConsistentC
  • 55
  • 5

1 Answers1

0

I'm late to the game but maybe someone will need this answer:

In my case that erro meant that the variables for callbacks should be defined outside the model building function so that they can be accessed by search.

In your particular case I think there might have been two possible causes:

  1. Callbacks should be given as a list - even if there is only one:

    callbacks = [early_stopping]

  2. Code formatting not according to PEP8: https://peps.python.org/pep-0008/

MarcinKamil
  • 135
  • 8