Make ColumnTransfomer automatically convert sparse inputs to CSR format
#30543
Conversation
Enhanced _check_X to check for sparse matrix subscriptability via __getitem__. Non-subscriptable sparse formats (e.g., COO, DIA) now raise a warning and are converted to CSR format for compatibility.
Enhanced _check_X to check for sparse matrix subscriptability via __getitem__. Non-subscriptable sparse formats (e.g., COO, DIA) now raise a warning and are converted to CSR format for compatibility.
ColumnTransfomer automatically convert sparse inputs to CSR format
Removed the User warning added in this PR. Sparse inputs are automatically converted to the CSR format as this is currently the most efficient approach.
There was a problem hiding this comment.
Please add a non-regression test based on the code snippet of the original issue under sklearn/compose/tests/test_column_transformer.py to check that it's now possible to use the column transformer on dia or COO formatted sparse matrix inputs without getting an error.
Please also add a changelog entry to document this fix under doc/whats_new/upcoming_changes/sklearn.compose.
1- removed Docstring from `_check_X`. 2- added a non-regression unit test for verification. 3- added the changelog entry.
|
any updates on this PR @ogrisel |
|
Hi @ogrisel, Thanks! |
|
Hi @thomasjpfan , |
|
hi @thomasjpfan i guess @ogrisel is not active can you assign someone else |
Fixes #30275
What does this implement/fix
Enhanced
_check_Xto verify for sparse matrix subscriptability via__getitem__. Non-subscriptable sparse formats now raise a warning and are converted to CSR format for compatibility.Any other comments?
The
_check_Xfunction insklearn/compose/_column_transformer.py, line 1324:Current behavior:
Only checks for
sparse.issparse(X)which returns True for dia_array(or any other sparse matrix type from scipy.sparse).This causes a
TypeError: 'dia_array' object is not subscriptablewhen non-subscriptable sparse formats are used andfit_transformis called.Fix:
This update ensures that non-subscriptable sparse formats are handled by raising a warning and converting them to CSR format, preventing errors during transformation.