You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
dm.scaffold.fuzzy_scaffolding is quite a powerful function but its output is often hard to understand and also process for downstream task.
We could keep backward compat by keeping dm.scaffold.fuzzy_scaffolding and propose an alternative function that will do the same kind of processing under the hood but return a data structure that is more intuitive and easier to use (a dataframe or a list of dataframe?).
The text was updated successfully, but these errors were encountered:
@hadim Ill try to take a jab at this. Let you know what happens. Question: Does the output results have multi valued attributes? I don't know much about the output as of yet, but this dataframe route could make it difficult to implement if there are multiple and/or range of data points for a single resulting input. Thanks.
Question: Does the output results have multi valued attributes?
Currently the output is a tuple of list and dicts. I would say for the refactoring, it's perfectly ok to completely rethink the output. Whether a dataframe would be a good fit would need to be assessed, but it's probably fine to have some new structure (a dataclass or something similar) to hold the results. It's likely that the code and output can be simplified too.
Happy to help around that if you are taking a jab at this.
dm.scaffold.fuzzy_scaffolding
is quite a powerful function but its output is often hard to understand and also process for downstream task.We could keep backward compat by keeping
dm.scaffold.fuzzy_scaffolding
and propose an alternative function that will do the same kind of processing under the hood but return a data structure that is more intuitive and easier to use (a dataframe or a list of dataframe?).The text was updated successfully, but these errors were encountered: