[Breaking Change] MapieRegressor removed in v1.x with no migration guidance
Summary
MapieRegressor was removed in MAPIE 1.x without a compatibility shim, deprecation warning, or official migration pathway for users upgrading from MAPIE 0.x tutorials and notebooks.
Any user following pre-1.0 MAPIE examples immediately encounters:
from mapie.regression import MapieRegressor
which now raises:
ImportError: cannot import name 'MapieRegressor'
This creates significant onboarding friction for practitioners migrating from older notebooks, Kaggle examples, blog posts, and StackOverflow solutions.
While debugging this in a Kaggle + Python 3.12 environment, I confirmed that MAPIE 1.x introduced a complete architectural redesign replacing MapieRegressor with specialized conformal regression classes.
Environment
Python: 3.12.x
MAPIE: 1.4.0
scikit-learn: 1.6.1
Environment: Kaggle Notebook / Linux
Steps to Reproduce
pip install mapie
from mapie.regression import MapieRegressor
Expected Behavior
Either:
MapieRegressor should exist as a compatibility shim emitting a DeprecationWarning
OR
The ImportError should guide users toward the new replacement classes such as:
OR
Official migration documentation should clearly explain:
Actual Behavior
ImportError: cannot import name 'MapieRegressor' from 'mapie.regression'
No migration guidance or replacement suggestions are provided.
Suggested Fix
Possible improvements:
Add a lightweight compatibility shim for MapieRegressor
Raise a custom migration-aware error message
Add a dedicated 0.x → 1.x migration guide
Add notebook migration examples for Kaggle/Colab users
Example compatibility guidance:
| MAPIE 0.x |
MAPIE 1.x |
| MapieRegressor(method='plus') |
CrossConformalRegressor |
| MapieRegressor(method='base') |
SplitConformalRegressor |
| method='quantile' |
ConformalizedQuantileRegressor |
I'm also happy to contribute documentation or a migration guide PR if maintainers think it would be useful.
[Breaking Change] MapieRegressor removed in v1.x with no migration guidance
Summary
MapieRegressorwas removed in MAPIE 1.x without a compatibility shim, deprecation warning, or official migration pathway for users upgrading from MAPIE 0.x tutorials and notebooks.Any user following pre-1.0 MAPIE examples immediately encounters:
which now raises:
This creates significant onboarding friction for practitioners migrating from older notebooks, Kaggle examples, blog posts, and StackOverflow solutions.
While debugging this in a Kaggle + Python 3.12 environment, I confirmed that MAPIE 1.x introduced a complete architectural redesign replacing
MapieRegressorwith specialized conformal regression classes.Environment
Steps to Reproduce
Expected Behavior
Either:
MapieRegressorshould exist as a compatibility shim emitting aDeprecationWarningOR
The ImportError should guide users toward the new replacement classes such as:
CrossConformalRegressorSplitConformalRegressorConformalizedQuantileRegressorOR
Official migration documentation should clearly explain:
old → new class mappings
predict() signature changes
alpha → confidence_level migration
Actual Behavior
No migration guidance or replacement suggestions are provided.
Suggested Fix
Possible improvements:
Add a lightweight compatibility shim for
MapieRegressorRaise a custom migration-aware error message
Add a dedicated 0.x → 1.x migration guide
Add notebook migration examples for Kaggle/Colab users
Example compatibility guidance:
I'm also happy to contribute documentation or a migration guide PR if maintainers think it would be useful.