This repository was archived by the owner on Feb 28, 2024. It is now read-only.
This repository was archived by the owner on Feb 28, 2024. It is now read-only.
Using Integer search space dimension with Tensorflow #790
Description
Hello,
consider the following minimal example, where I am optimizing one hyperparameter of a dummy neural network:
space = [Integer(2, 4, name='kernel_size')]
@use_named_args(space)
def objective_function(kernel_size):
model = keras.Sequential()
model.add(Convolution1D(filters=64, kernel_size=kernel_size))
model.compile(optimizer=Adam(), loss='binary_crossentropy',
metrics=['acc'])
return random.randint(0, 100)
res = gp_minimize(objective_function, space, n_calls=10, verbose=True)Running this with scikit-optimize==0.5.2, I get the following error:
Traceback (most recent call last):
File ".\venv\lib\site-packages\keras\utils\conv_utils.py", line 33, in normalize_tuple
value_tuple = tuple(value)
TypeError: 'numpy.int32' object is not iterable
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "./recreate.py", line 22, in <module>
res = gp_minimize(objective_function, space, n_calls=10, verbose=True)
File ".\venv\lib\site-packages\skopt\optimizer\gp.py", line 228, in gp_minimize
callback=callback, n_jobs=n_jobs)
File ".\venv\lib\site-packages\skopt\optimizer\base.py", line 248, in base_minimize
next_y = func(next_x)
File ".\venv\lib\site-packages\skopt\utils.py", line 636, in wrapper
objective_value = func(**arg_dict)
File "./recreate.py", line 16, in objective_function
model.add(Convolution1D(filters=64, kernel_size=first_convolution_layer_kernel_size))
File ".\venv\lib\site-packages\keras\legacy\interfaces.py", line 91, in wrapper
return func(*args, **kwargs)
File ".\venv\lib\site-packages\keras\layers\convolutional.py", line 353, in __init__
**kwargs)
File ".\venv\lib\site-packages\keras\layers\convolutional.py", line 109, in __init__
'kernel_size')
File ".\venv\lib\site-packages\keras\utils\conv_utils.py", line 36, in normalize_tuple
str(n) + ' integers. Received: ' + str(value))
ValueError: The `kernel_size` argument must be a tuple of 1 integers. Received: 3
This is due to this line value_tuple = tuple(value) in Keras which fails for numpy.int32 types.
One can resolve this by using model.add(Convolution1D(filters=64, kernel_size=kernel_size.item())) instead of model.add(Convolution1D(filters=64, kernel_size=kernel_size)). However this is not a clean solution as it restricts reusing the objective function with other types.
Is there a proper way to convert the numpy.int32 parameter to a native python type?