Issues

[BUG] model_config__* keys are silently ignored when the estimator was built without a model_config

OpenTabular/DeepTab#613 · opened

on Oct 21, 2026

[DOCS] Migration guide shows the v1 DefaultMambularConfig import from deeptab.models instead of deeptab.configs

OpenTabular/DeepTab#611 · opened

on Oct 21, 2026

[DOCS] ff_dropout is documented and tuned for FTTransformer/TabTransformer but has no effect there

OpenTabular/DeepTab#609 · opened

on Oct 21, 2026

[DOCS] sklearn_api.md calls nested set_params keys on a default-constructed estimator, which raises ValueError

OpenTabular/DeepTab#607 · opened

on Oct 21, 2026

[DOCS] README observability example does not produce lifecycle.jsonl or tensorboard/ shown in its run tree

OpenTabular/DeepTab#605 · opened

on Oct 21, 2026

[DOCS] Classifier pretrain() docstring says regression mode is enabled; it runs with regression=False

OpenTabular/DeepTab#603 · opened

on Oct 21, 2026

[DOCS] deeptab.models.experimental docstring imports ModernNCA, Tangos and Trompt, which the package does not export

OpenTabular/DeepTab#601 · opened

on Oct 21, 2026

[DOCS] DeepTabMetric.higher_is_better is described as driving HPO and early stopping, but nothing reads it

OpenTabular/DeepTab#599 · opened

on Oct 21, 2026

[DOCS] Task-metadata examples show num_classes=2 for a binary classifier; the artifact stores 1 (output width)

OpenTabular/DeepTab#597 · opened

on Oct 21, 2026

[DOCS] ObservabilityConfig.tensorboard_name is documented as the TensorBoard sub-directory but is never used

OpenTabular/DeepTab#595 · opened

on Oct 21, 2026

[DOCS] TaskModel docstring examples import deeptab.models.MLP and deeptab.architectures.ResNetModel, which do not exist

OpenTabular/DeepTab#593 · opened

on Oct 21, 2026

[DOCS] Installation troubleshooting snippet uses FTTransformerClassifier without importing it

OpenTabular/DeepTab#591 · opened

on Oct 21, 2026

[DOCS] Metric examples call LSS fit() without the required family argument

OpenTabular/DeepTab#589 · opened

on Oct 21, 2026

[DOCS] Pretraining docs misstate which models support pretrain()

OpenTabular/DeepTab#587 · opened

on Oct 21, 2026

[DOCS] Skewed-regression tutorial recommends numerical_method="splines", which does not exist

OpenTabular/DeepTab#585 · opened

on Oct 21, 2026

[DOCS] Upgrade notes call the v1 Mambular* data-module names deprecated, but they were removed

OpenTabular/DeepTab#583 · opened

on Oct 21, 2026

[DOCS] Homepage says losses are selectable through TrainerConfig, which has no loss field

OpenTabular/DeepTab#581 · opened

on Oct 21, 2026

[DOCS] Observability tutorial uses default_root_dir, which does not move the checkpoint of a plain fit()

OpenTabular/DeepTab#579 · opened

on Oct 21, 2026

[BUG] fit() crashes with BatchNorm when the training rows leave a one-row last batch (e.g. 162 rows)

OpenTabular/DeepTab#577 · opened

on Oct 21, 2026

[DOCS] Config JSON round-trip example fails: model configs hold nn.Module activations

OpenTabular/DeepTab#575 · opened

on Oct 21, 2026

[DOCS] TabularBatch loop example never converts loader batches (they are lists, not tuples)

OpenTabular/DeepTab#573 · opened

on Oct 21, 2026

[DOCS] Quickstart says numerical features are standardized by default, but the default is min-max scaling followed by PLE

OpenTabular/DeepTab#571 · opened

on Oct 21, 2026

[DOCS] Quickstart examples fail when run in order (overwritten split, "..." placeholder)

OpenTabular/DeepTab#569 · opened

on Oct 21, 2026

[BUG] load() across the LSS boundary builds a broken estimator (e.g. MLPRegressor.load on an LSS artifact)

OpenTabular/DeepTab#567 · opened

on Oct 21, 2026

[DOCS] Run-directory trees name the checkpoint best.ckpt, but fit() writes best_model.ckpt

OpenTabular/DeepTab#565 · opened

on Oct 21, 2026

Navigation Menu

Repository navigation

Issues

[BUG] model_config__* keys are silently ignored when the estimator was built without a model_config

[DOCS] Migration guide shows the v1 DefaultMambularConfig import from deeptab.models instead of deeptab.configs

[DOCS] ff_dropout is documented and tuned for FTTransformer/TabTransformer but has no effect there

[DOCS] sklearn_api.md calls nested set_params keys on a default-constructed estimator, which raises ValueError

[DOCS] README observability example does not produce lifecycle.jsonl or tensorboard/ shown in its run tree

[DOCS] Classifier pretrain() docstring says regression mode is enabled; it runs with regression=False

[DOCS] deeptab.models.experimental docstring imports ModernNCA, Tangos and Trompt, which the package does not export

[DOCS] DeepTabMetric.higher_is_better is described as driving HPO and early stopping, but nothing reads it

[DOCS] Task-metadata examples show num_classes=2 for a binary classifier; the artifact stores 1 (output width)

[DOCS] ObservabilityConfig.tensorboard_name is documented as the TensorBoard sub-directory but is never used

[DOCS] TaskModel docstring examples import deeptab.models.MLP and deeptab.architectures.ResNetModel, which do not exist

[DOCS] Installation troubleshooting snippet uses FTTransformerClassifier without importing it

[DOCS] Metric examples call LSS fit() without the required family argument

[DOCS] Pretraining docs misstate which models support pretrain()

[DOCS] Skewed-regression tutorial recommends numerical_method="splines", which does not exist

[DOCS] Upgrade notes call the v1 Mambular* data-module names deprecated, but they were removed

[DOCS] Homepage says losses are selectable through TrainerConfig, which has no loss field

[DOCS] Observability tutorial uses default_root_dir, which does not move the checkpoint of a plain fit()

[BUG] fit() crashes with BatchNorm when the training rows leave a one-row last batch (e.g. 162 rows)

[DOCS] Config JSON round-trip example fails: model configs hold nn.Module activations

[DOCS] TabularBatch loop example never converts loader batches (they are lists, not tuples)

[DOCS] Quickstart says numerical features are standardized by default, but the default is min-max scaling followed by PLE

[DOCS] Quickstart examples fail when run in order (overwritten split, "..." placeholder)

[BUG] load() across the LSS boundary builds a broken estimator (e.g. MLPRegressor.load on an LSS artifact)

[DOCS] Run-directory trees name the checkpoint best.ckpt, but fit() writes best_model.ckpt