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
{{ message }}
Repository navigation
[ENH] list_estimation_procedures() should include the procedure IDs #1750
Within evaluations/functions.py/list_estimation_procedures an API call is made to obtain all the existing estimation procedures, the response contains an ID response as well as a TaskTypeID which could prove useful to know when a user is trying to create a new task.
The current implementation only returns the procedure name, and the documentation is a bit lacking.
I would suggest making the return a dictionary by default with the procedure ID as the key to an inner dict containing other information, while allowing for an optional output format parameter to select pandas DataFrame.
@L-Agbaria
I’ll inspect the current return format and the API response, then update list_estimation_procedures() to expose the procedure ID and related information without losing the existing behaviour where possible. I’ll also add tests and update the documentation.
Does that match the intended direction?
Before a merge is made, let us test if the code functions normally for all cases of loading tasks and creating runs/flows. Those are things that might bypass the automatic checks.
Activity
Jayant-kernel commentedon Sep 23, 2026
@L-Agbaria
I’ll inspect the current return format and the API response, then update
list_estimation_procedures()to expose the procedure ID and related information without losing the existing behaviour where possible. I’ll also add tests and update the documentation.Does that match the intended direction?
L-Agbaria commentedon Sep 25, 2026
Before a merge is made, let us test if the code functions normally for all cases of loading tasks and creating runs/flows. Those are things that might bypass the automatic checks.
Jayant-kernel commentedon Sep 25, 2026
Okay I’ll test task loading and a basic run/flow creation before opening the PR, and include the results there.