Last meeting we discussed giving RMSE & Huber Loss a go on CGG data.
We currently don't support Huber loss in torchdms, but PyTorch has an implementation that will be quick and easy to integrate.
I'll begin by just using the default parameters for delta, which results in a SmoothL1Loss -- but will add some capabilities to toggle this after things are up and running.
Last meeting we discussed giving RMSE & Huber Loss a go on CGG data.
We currently don't support Huber loss in
torchdms, but PyTorch has an implementation that will be quick and easy to integrate.I'll begin by just using the default parameters for delta, which results in a
SmoothL1Loss-- but will add some capabilities to toggle this after things are up and running.