Fix ImportError with sklearn 1.7.0+ by replacing tosequence

[AddyM]

Description

Fixes #267

This PR resolves the ImportError: cannot import name 'tosequence' from 'sklearn.utils' that occurs with scikit-learn >= 1.7.0.

Root Cause

The tosequence utility function was deprecated in scikit-learn 1.5 and completely removed in version 1.7.0, breaking sklearn-pandas on import.

Changes Made

In sklearn_pandas/pipeline.py:

• Removed the import of sklearn.utils.tosequence
• Added steps = list(steps) at the beginning of TransformerPipeline.__init__
• Changed self.steps = tosequence(steps) to self.steps = steps

In tests/test_pipeline.py:

• Added regression tests for list input
• Added regression tests for tuple input
• Added regression tests for generator input
• Added test verifying steps is always a list type

Why This Works

The tosequence utility simply converted various input types to a list. By calling list(steps) at the start of __init__, we:

• Provide identical functionality to the original tosequence
• Handle generators/iterators correctly (consumed early, before zip())
• Remove dependency on sklearn internal utilities
• Work across all sklearn versions
• Simplify the codebase

Testing

• All existing tests pass (4/4 original tests)
• All new regression tests pass (4/4 new tests)
• Manually tested with sklearn 1.7.2
• Verified backward compatibility
• Tested with list, tuple, and generator inputs

Test Results

8 passed in 0.76s

Compatibility

• scikit-learn >= 1.7.0 (broken before this fix)
• scikit-learn 1.5.x - 1.6.x (works before and after)
• scikit-learn < 1.5.0 (works before and after)

Checklist

• Code changes implement the fix
• Tests added for regression prevention
• All tests passing
• No breaking changes to public API
• Backward compatible with older sklearn versions