favyen2
commented
8 months ago There is config object now but I think there are still some things to look into:
- We should consider using YAML for dataset configuration so it's more readable. Since this config will mostly be manually edited.
- Right now the user has to specify everything in the config file like bands and layer type even though some of that can come from the data source. We should consider compiling the YAML into a JSON like the current config format where missing values are automatically populated.
Figure out how configuration files and loading config from them should work.
Right now some configuration loads arguments to a constructor directly (e.g. loading data source), while others have dedicated parsing functions.
Look into best practices but I think sticking with a function to parse the configuration into a class is good.
Dataset should store config as DatasetConfig or similar rather than keeping the decoded JSON dict.