I am using FastAI v2 to train a model with WandCallback. I want to run an experimentation to assess which loss function combination would yield the best model? So, I have losses like:
losses = [‘Dice’, ‘Focal’, ‘Tversky’, ‘BCE’]
alphas = [.1,.2,.3,.4,.5,.6,.7,.8,.9]
And I have a combo loss CustomComboLoss class which accepts L1 and L2 along with alpha as arguments to generate a loss function for a run.
I don’t want to run it manually. Can this be performed using WandCallback and sweep?
Any suggestions or links to useful resources will be very much appreciated.
Many Thanks and