Open gregtatum opened 4 months ago
It's similar to #599 but I disagree that we should put Taskcluster-specific logic into the pipeline training script. The design so far was to make everything inside the pipeline
directory independent of an orchestrator. It also makes it more testable. I think having one wrapper for Taskcluster specific needs is fine. We also already have #579. So if you agree on the current separation of concerns we can close this one.
I haven't commented on this one as I think it's worth looking at both approaches and come to a consensus on the design. In #842 I'm rewriting the train.sh to train.py, but the taskcluster bits were too complicated to initially take on.
We have:
It will be much simpler to put this into a single training script. We should probably do this after our first big training run.