Would it be wise to first allow each GPU to train its own version of the model independently on it's given data sunset and cross validate using the rest of the data set?
I'm thinking this would be a sanity check of sorts to see if the chosen model is worth further training.
Would it be wise to first allow each GPU to train its own version of the model independently on it's given data sunset and cross validate using the rest of the data set?
I'm thinking this would be a sanity check of sorts to see if the chosen model is worth further training.