Move application functions to own module?

[nsfabina]

Seems like data_management is for building data and handling samples, primarily?

If so, maybe an application module for taking a completed model and applying it to any input raster?

[pgbrodrick]

I'm not totally convinced I want to do this....this is principally the reciprocal of the data build problem, though slightly different. If you feel strongly about having it as it's own module, or there's a good argument for it, it's not hard to do...but right now I kindof feel like it's in the right spot.

[nsfabina]

I don't know if I really see "take trained model and apply to data rasters" as the inverse/reciprocal operation to "take data rasters and convert them into a specific file/data format", in the same way that I see "read data from file" as the inverse/reciprocal to "write data to file". Maybe there would be more of a case for this if the two tasks were sharing a fair amount of code, but it would still be possible to factor out that code as general utility functions, like "read" or "write" operations.

High level, this comes down to clarity in naming and organization for me, since it's possible to put this code anywhere in the project and have it function as intended. We've been talking a lot about making this project intuitive to users that may not be Python or deep learning experts, and breaking up the modules based on research/scientific task seems to be one of those choices:

• clean/preprocess data (this is new, haven't had a chance to look into it yet)
• build data
• train model with data
• generate reports on built data or trained models

To me, it seems intuitive to add another module for that task-based framework:

• apply trained model to data

Aside from the task-based framework, there's the naming side of things. It makes sense to me that building data, transforming/scaling data, or creating data generators are all in data_management. Similarly, I wouldn't find any experimentation or reporting functionality in data_management because those are distinct areas. Where do I look for code that applies models to data? data_management doesn't seem like an obvious place for that because we're not really managing data, right?

Please let me know if you have other thoughts on the framework we're using to organize the project or if you're looking at this from a different perspective.