Issues

Representations registered with supervision="optional" never receive the Preprocessor's target_aware, so target_aware=False still fits on y

OpenTabular/PreTab#96 · opened

on Oct 9, 2026

A column with no observed values at fit crashes the Preprocessor (numerical, one-hot) or silently disappears from the output (int)

OpenTabular/PreTab#95 · opened

on Oct 9, 2026

Preprocessor rejects polars DataFrames, so pipelines with set_output(transform="polars") crash

OpenTabular/PreTab#94 · opened

on Oct 9, 2026

An invalid task such as "Regression" is silently treated as classification by the splines, the location selectors and the presets

OpenTabular/PreTab#93 · opened

on Oct 9, 2026

RepresentationSearchCV around a classifier exposes no predict_proba / classes_ and is not a classifier, so nested CV with roc_auc returns NaN

OpenTabular/PreTab#92 · opened

on Oct 9, 2026

RepresentationSearchCV cannot use group-based CV splitters and accepts no groups or fit parameters

OpenTabular/PreTab#91 · opened

on Oct 9, 2026

CrossFittedTransformer.fit_transform forces a dense float64 output: string columns crash, sparse output and dtype differ from transform

OpenTabular/PreTab#90 · opened

on Oct 9, 2026

CrossFittedTransformer(Preprocessor(add_missing_indicator=True)) crashes when a fold's training rows contain no missing value

OpenTabular/PreTab#89 · opened

on Oct 9, 2026

B/M/I spline transformers crash on a 2-D y (multi-output targets) even when they do not use the target

OpenTabular/PreTab#88 · opened

on Oct 9, 2026

BSpline / MSpline / ISpline transformers encode a missing value as an all-zero basis row; for the I-spline it equals the training minimum

OpenTabular/PreTab#87 · opened

on Oct 9, 2026

Standalone numerical transformers drop DataFrame feature names: reordered columns are silently accepted and outputs are named x0_...

OpenTabular/PreTab#85 · opened

on Oct 9, 2026

A partial policy mapping switches clip-family splines to extrapolate, and extrapolate/warn turn out-of-range rows into all-zero basis rows

OpenTabular/PreTab#84 · opened

on Oct 9, 2026

Unsupervised ReLUExpansionTransformer places its last center at the training maximum, producing a dead column on every fit

OpenTabular/PreTab#83 · opened

on Oct 9, 2026

Unsupervised quantile placement repeats feature-map centers on tied data, producing duplicate RBF/ReLU/sigmoid/tanh columns

OpenTabular/PreTab#82 · opened

on Oct 9, 2026

Spline knot_locations is still sized by output_dim: other knot counts raise, adaptive trims them, out-of-range knots are silently replaced

OpenTabular/PreTab#81 · opened

on Oct 9, 2026

Default to_spec() files contain bare NaN tokens and are not valid JSON for strict parsers

OpenTabular/PreTab#80 · opened

on Oct 9, 2026

CrossFittedTransformer(Preprocessor) refits column types and category codes per fold, so out-of-fold features use a different encoding than transform

OpenTabular/PreTab#79 · opened

on Oct 9, 2026

Preprocessor fitted on a DataFrame rejects a same-width NumPy array at transform (and vice versa)

OpenTabular/PreTab#78 · opened

on Oct 9, 2026

RepresentationSearchCV with a classifier keeps task="regression": string labels crash and target-aware placement uses the wrong task

OpenTabular/PreTab#77 · opened

on Oct 9, 2026

Preprocessor with one-hot encoding (and preset="expanded") crashes under sklearn.set_config(transform_output="pandas")

OpenTabular/PreTab#76 · opened

on Oct 9, 2026

output_format="auto" chooses sparse vs dense per transform batch, breaking dense-only estimators at predict

OpenTabular/PreTab#73 · opened

on Oct 2, 2026

Preprocessor.transform does not check n_features_in_ for ndarray input; extra columns are silently misaligned

OpenTabular/PreTab#72 · opened

on Oct 2, 2026

Preprocessor.fit crashes when a column name starts with '_' or contains '__'

OpenTabular/PreTab#71 · opened

on Oct 2, 2026

CART location selector can drop the dominant split because spacing is enforced in ascending location order

OpenTabular/PreTab#70 · opened

on Oct 2, 2026

Target-aware spline placement trims selector knots by position, discarding the most informative splits; adaptive width capped at 15

OpenTabular/PreTab#69 · opened

on Oct 2, 2026

Navigation Menu

Repository navigation

Issues

Representations registered with supervision="optional" never receive the Preprocessor's target_aware, so target_aware=False still fits on y

A column with no observed values at fit crashes the Preprocessor (numerical, one-hot) or silently disappears from the output (int)

Preprocessor rejects polars DataFrames, so pipelines with set_output(transform="polars") crash

An invalid task such as "Regression" is silently treated as classification by the splines, the location selectors and the presets

RepresentationSearchCV around a classifier exposes no predict_proba / classes_ and is not a classifier, so nested CV with roc_auc returns NaN

RepresentationSearchCV cannot use group-based CV splitters and accepts no groups or fit parameters

CrossFittedTransformer.fit_transform forces a dense float64 output: string columns crash, sparse output and dtype differ from transform

CrossFittedTransformer(Preprocessor(add_missing_indicator=True)) crashes when a fold's training rows contain no missing value

B/M/I spline transformers crash on a 2-D y (multi-output targets) even when they do not use the target

BSpline / MSpline / ISpline transformers encode a missing value as an all-zero basis row; for the I-spline it equals the training minimum

Standalone numerical transformers drop DataFrame feature names: reordered columns are silently accepted and outputs are named x0_...

A partial policy mapping switches clip-family splines to extrapolate, and extrapolate/warn turn out-of-range rows into all-zero basis rows

Unsupervised ReLUExpansionTransformer places its last center at the training maximum, producing a dead column on every fit

Unsupervised quantile placement repeats feature-map centers on tied data, producing duplicate RBF/ReLU/sigmoid/tanh columns

Spline knot_locations is still sized by output_dim: other knot counts raise, adaptive trims them, out-of-range knots are silently replaced

Default to_spec() files contain bare NaN tokens and are not valid JSON for strict parsers

CrossFittedTransformer(Preprocessor) refits column types and category codes per fold, so out-of-fold features use a different encoding than transform

Preprocessor fitted on a DataFrame rejects a same-width NumPy array at transform (and vice versa)

RepresentationSearchCV with a classifier keeps task="regression": string labels crash and target-aware placement uses the wrong task

Preprocessor with one-hot encoding (and preset="expanded") crashes under sklearn.set_config(transform_output="pandas")

output_format="auto" chooses sparse vs dense per transform batch, breaking dense-only estimators at predict

Preprocessor.transform does not check n_features_in_ for ndarray input; extra columns are silently misaligned

Preprocessor.fit crashes when a column name starts with '_' or contains '__'

CART location selector can drop the dominant split because spacing is enforced in ascending location order

Target-aware spline placement trims selector knots by position, discarding the most informative splits; adaptive width capped at 15