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
[BUG] Error: No extension registered which can handle flow when using 'get_flow()' with reinstantiate=True #1626
When trying to use get_flow(1) with reinstantiate=True the following error is raised: ValueError: No extension registered which can handle flow: 1 (openml.evaluation.EuclideanDistance).
It's the same with all the different flow ids I have tried.
To reproduce
from openml.flows.functions import get_flow
flow = get_flow(1, reinstantiate=True)
you can try a different flow from catalogue which uses sklearn, but that will lead to another problem that all the sklearn tagged flows I see there are bounded to version 0.20.x or less
We have already discussed about decoupling the concept of estimator instansces from versions - but maybe we also should consider this for flows somehow?
you can try a different flow from catalogue which uses sklearn, but that will lead to another problem that all the sklearn tagged flows I see there are bounded to version 0.20.x or less
Yup, now I got ValueError: Trying to deserialize a model with dependency sklearn==0.18.1 not satisfied.
What to do then?
Yup, now I got ValueError: Trying to deserialize a model with dependency sklearn==0.18.1 not satisfied.
What to do then?
not really a bug, flows depend on library versions, so they expect the setup to have correct versions, does it affect you from running a test? can you tell which test-case exactly troubles you?
Hi @EmanAbdelhaleem !
I've worked on this issue and identified that the error occurs for OpenML-native flows (like evaluation functions) that don't have registered extensions. I've implemented a fix that makes get_flow() gracefully handle such flows by logging a warning instead of raising an error when reinstantiate=True is passed but no extension is available.
The version mismatch issue mentioned in the comments (sklearn==0.18.1 not satisfied) is expected behavior - flows depend on specific library versions. Users can use strict_version=False to work around version mismatches.
I'll submit a PR shortly to fix the original issue.
Activity
geetu040 commentedon Jan 22, 2026
you can try a different flow from catalogue which uses sklearn, but that will lead to another problem that all the sklearn tagged flows I see there are bounded to version
0.20.xor lessfkiraly commentedon Jan 22, 2026
these are ancient versions.
We have already discussed about decoupling the concept of estimator instansces from versions - but maybe we also should consider this for flows somehow?
EmanAbdelhaleem commentedon Jan 28, 2026
Yup, now I got
ValueError: Trying to deserialize a model with dependency sklearn==0.18.1 not satisfied.What to do then?
geetu040 commentedon Jan 30, 2026
not really a bug, flows depend on library versions, so they expect the setup to have correct versions, does it affect you from running a test? can you tell which test-case exactly troubles you?
Fix get_flow() with reinstantiate=True for flows without extensions
ssudhiravinesh commentedon Mar 7, 2026
Hi @EmanAbdelhaleem !
I've worked on this issue and identified that the error occurs for OpenML-native flows (like evaluation functions) that don't have registered extensions. I've implemented a fix that makes
get_flow()gracefully handle such flows by logging a warning instead of raising an error whenreinstantiate=Trueis passed but no extension is available.The version mismatch issue mentioned in the comments (sklearn==0.18.1 not satisfied) is expected behavior - flows depend on specific library versions. Users can use
strict_version=Falseto work around version mismatches.I'll submit a PR shortly to fix the original issue.