Our three ML tasks (dynamic flood simulation, GeoAI segmentation, and tile2net segmentation) have different, highly-customized implementations. We should try to create one interface for ML tasks that will allow for the addition of future tasks without custom implementation. We explored the idea of using a JSON specification for defining tasks that exist in a python module. Each task should specify what inputs it requires (including UVDAT types like Dataset, VectorData, Chart, etc.), what outputs it creates, package/system requirements, and how to invoke a main function.
This issue will require further discussion and planning prior to any implementation.
Our three ML tasks (dynamic flood simulation, GeoAI segmentation, and tile2net segmentation) have different, highly-customized implementations. We should try to create one interface for ML tasks that will allow for the addition of future tasks without custom implementation. We explored the idea of using a JSON specification for defining tasks that exist in a python module. Each task should specify what inputs it requires (including UVDAT types like Dataset, VectorData, Chart, etc.), what outputs it creates, package/system requirements, and how to invoke a main function.
This issue will require further discussion and planning prior to any implementation.