I have a bit of a conceptual question regarding reuse of assets/ops between model training/evaluation and inference and the docs seem to only discuss one or other use case.
The sensors example seems to create runs for individual files, which seems appropriate for inference. Then all the asset examples seem to load entire datasets, which seems appropriate for model training/evaluation. But how can you reuse ops in both cases? Is it better to write ops for datasets and create datasets of one from sensors, or is there someway to map an operation of a single entry onto a dataset for training/evaluation use cases?
Thanks in advance for the help