Review and comments data loaders

[fkiraly]

Short review of data loaders, with the aim to ensure easy useability by other project members and maintainability.

Referring to current state of 1.01-cjr-change-point-index-creator.ipynb, top cell

Main comments:

• first, and importantly, I would move the data loading code out of the jupyter notebook and into py files, possibly somewhere in the data folder. I would also organize the repository that all data concerns are separate, possibly also separating processed data from raw data. (more generally, raw data above a certain total size, perhaps 10 MB, should not be in GitHub repos unless dedicated data repos, as it clutters the repo)
• the jupyter notebook would then just import from the "mini-package".
• once the data loader is in the py file, I would strongly recommend to refactor it. At the moment it is a monolithic end-to-end function with subroutines inside. I would suggest to have a design that separates: (a) switching between file and in-memory, (b) processing pipeline; the processing is also a loop over files, so I would make it "process a single file". More precisely, I would suggest to split things as follows:
  • raw data loader: file to memory, for raw data to in-memory repr of raw data
  • processed data loader: file to memory, for processed data
  • processed data saver: memory to file, for processed data
  • data processing: memory in, memory out, for a single file
  • primary user facing routines with syntactic sugar:
    • processed data loader with sugar: if processed files not present, runs everything until those are created, and makes default choices for locations. Also loops over multiple files. Optionally, one could force rerun and do checksums.
    • raw data loader with sugar
    • data processing for multiple files sugar. Could be the same method as above, but still might make sense to add the sugar in a second step to avoid.
• finally, the different methods should have correct docstrings. The docstrings should reference the data dictionaries, see #108, so users know what format they can expect the data in.
• optional, but recommended best practice: tests for the loaders. There are already test files in the repo, so we could use them for tests via pytest and pydantic or similar.

[RobotPsychologist]

Thanks for the feedback @fkiraly!

1. My intention was to move the function in some form into this file here once I had demonstrated it is working in the notebook: 0_meal_identification/meal_identification/meal_identification/dataset.py

So, instead of having a dataset.py script for each project, do I have one for all projects in the bg_control/data folder? I was initially planning on having the script with classes and functions available to use in 0_meal_identification/meal_identification/meal_identification/dataset.py (with a similar file for each project) , and the data was going to be stored in:

• 0_meal_identification/meal_identification/data/raw
• 0_meal_identification/meal_identification/data/processed
• 0_meal_identification/meal_identification/data/interim
• 0_meal_identification/meal_identification/data/external

[RobotPsychologist]

@andytubeee if you're interested.

[RobotPsychologist]

@andytubeee @Phiruby @Tony911029 please check if there is anything that Franz mentioned here that might also be included in the data work as an enhancement.