Include `pool_size` as hyperparameter to optimise over
tallamjr opened this issue · 1 comments
tallamjr commented
Currently the pool_size
configuration in some layers are set to the self.kernel_size
as this was previously the same setting as 3 for both.
It might be worth investigating the effect of allowing this to vary and to be optimised for.
The code for adding something like this could look like:
pool_size = trial.suggest_categorical("kernel_size", [3, 6, 12, 24, 48, 96]) # --> Pooling width
...
model = SNXModel(
num_classes=num_classes,
kernel_size=kernel_size,
pool_size=pool_size
)
With modifications made to snX/layers.py
file accordingly