search

nhsengland / NHSSynth

Package to accompany P41
https://nhsengland.github.io/NHSSynth/
MIT License
4 stars 1 forks source link

Better way to propagate generations and evaluations than the current nested dictionaries? #85

Closed HarrisonWilde closed 1 year ago

HarrisonWilde commented 1 year ago

The design feels overly complex to account for the case where a user does not want to provide a seed (which is perhaps a degenerative case anyway).

We could:

  1. Remove the option of no seed, i.e. default the seed to some value if it isn't specified. Then we could potentially store everything in a nested dictionary (it is still a little annoying to have to unpack the nested dict though).
  2. Use a dataframe to carry through results, i.e. experiment_results: pd.DataFrame contains a column called data, model, training_metrics, and is appended with the outcomes of the evaluation module? Feels weird to nest dataframes though.
HarrisonWilde commented 1 year ago

Solved by the reworking of the EvalFrame.