[ChrisW09]

Fixes #72.

A Preprocessor fitted on a NumPy array accepted a transform-time array with extra columns without an error. A leading extra column (for example an accidental ID column) shifted every feature and dropped the last one, and a trailing extra column was ignored, so downstream predictions were silently wrong.

Root cause

Preprocessor.transform never compared the input width with n_features_in_. Array columns are named feature_0, feature_1, ... by position, and the inner ColumnTransformer, fitted on a DataFrame, selects those names and ignores any further columns. An array with extra columns therefore still contained every fitted name. A narrower array raised only an indirect columns are missing error.

Fix

Preprocessor.transform checks 2-D NumPy input before converting it to a DataFrame. If the number of columns differs from n_features_in_, it raises scikit-learn's usual X has n features, but Preprocessor is expecting m features as input. as a PretabDataError, which is a ValueError. DataFrame and dict input is still matched by column name.

Behaviour changes

• transform with a NumPy array of a different width than at fit now raises PretabDataError with the scikit-learn message. Before, a wider array was accepted and misaligned, and a narrower one raised ValueError: columns are missing: {...}. Code that catches ValueError still catches the new error.

Documentation

• Preprocessor.transform docstring: new Raises section for an array whose width differs from n_features_in_.
• docs/core_concepts/configuration.md ("NumPy array input"): array columns are matched by position, so transform raises the usual scikit-learn error for an array with a different number of columns.

Tests

• test_transform_rejects_an_ndarray_of_a_different_width[2], [4]: a narrower and a wider array both raise the scikit-learn-style message. Fails on main (width 2 raises columns are missing, width 4 is accepted), passes here.
• test_transform_accepts_an_ndarray_of_the_fitted_width: an array of the fitted width is still accepted. Guard test: passes on both by design.

Reproduction

Output of the script from #72 on this branch:

n_features_in_: 3
predict(X)    : [222.66 100.1  199.4 ]
predict(X_bad): PretabDataError: X has 4 features, but Preprocessor is expecting 3 features as input.
ValueError: X has 4 features, but StandardScaler is expecting 3 features as input.

Checks

• just test: 1441 passed, 59 skipped, 7 xfailed; coverage 93.65% (gate 90%).
• Minimum dependencies (CI job: Python 3.10, numpy 1.24.4, pandas 2.0.3, scipy 1.10.1, scikit-learn 1.6.0): 1428 passed, 72 skipped, 7 xfailed; scripts/quickstart.py passes.
• just docs: strict build succeeds.
• just check: all pre-commit and pre-push hooks pass (ruff format, ruff lint, prettier, file hygiene, pyright with 0 errors).
• Import direction between subpackages is unchanged.

Merge notes

• Base: main.
• The PR for Preprocessor fitted on a DataFrame rejects a same-width NumPy array at transform (and vice versa) #78 (fix(preprocessor): match array input by position after a named fit #116) moves the same width check into Preprocessor._align_input and extends it to a DataFrame passed after an array fit. pretab/preprocessor.py conflicts in transform, in both the inline check and the Raises text. Take Preprocessor fitted on a DataFrame rejects a same-width NumPy array at transform (and vice versa) #78's side (X = self._align_input(X) and its Raises text), which keeps this PR's behaviour. This PR's tests pass unchanged on the result, and the full suite passes (1444 passed, 59 skipped, 7 xfailed). docs/core_concepts/configuration.md merges without a conflict but then states the width error twice, so drop this PR's sentence when both land. tests/integration/test_preprocessor.py has an append/append conflict: keep both blocks.
• The PRs for Preprocessor.get_feature_names_out(input_features) fails after fitting on an ndarray, breaking Pipeline.get_feature_names_out() #65 (fix(preprocessor): accept input_features after fitting on a NumPy array #108) and Preprocessor rejects polars DataFrames, so pipelines with set_output(transform="polars") crash #94 (fix(preprocessor): accept polars DataFrames as input #129) also append tests at the end of tests/integration/test_preprocessor.py, so expect an append/append conflict: keep both blocks.
• Follow-up with Preprocessor rejects polars DataFrames, so pipelines with set_output(transform="polars") crash #94: that PR reads a list of rows like an array. This check only looks at np.ndarray, so on a trial merge a list of rows with an extra column is still accepted after an array fit. When Preprocessor rejects polars DataFrames, so pipelines with set_output(transform="polars") crash #94 lands, the check should cover all input without column labels (Preprocessor rejects polars DataFrames, so pipelines with set_output(transform="polars") crash #94's has_column_labels), not only arrays.

This change was previously part of #75, which is closed in favour of one pull request per issue.