Would it be possible to make some changes to the architecture to allow preproccess chaining?
Id like to preprocess with algorithm A, potentially further preprocessing with algorithm B, and finally training with C.
A -> B -> C
From my perspective, this might be doable if /preprocessedData is writeable during training, but Im unsure how you guys are managing mounts and disk images.
Created by Clinton Mielke subcosmos Hi, thanks for this interesting suggestion. We don't feel a substantial change to the current processing pipeline is feasible at this stage in the challenge. If you feel that such a change would have a significant impact on participants' ability to perform well please elaborate. We'd be interested to hear your thoughts.