New tabpfn transformers #239
Conversation
There was a problem hiding this comment.
Pull request overview
This PR introduces three new TabPFN-based transformers for Driverless AI that leverage pre-trained TabPFN models for outlier detection and embedding generation. The implementation includes both transformer and model components for unsupervised outlier detection, along with a supervised embedding transformer.
Key Changes:
- Adds TabPFN-based outlier detection transformer with chunked processing and memory optimization
- Implements TabPFN embedding transformer using supervised learning with SVD dimensionality reduction
- Introduces unsupervised outlier detection model with Random Forest-based feature selection and density-aware sampling
Reviewed changes
Copilot reviewed 1 out of 1 changed files in this pull request and generated 10 comments.
| File | Description |
|---|---|
| transformers/outliers/tabpfn_outlier.py | Implements outlier detection transformer with chain-rule probability estimation across feature permutations, supporting chunked processing for large datasets |
| transformers/generic/tabpfn_embedding.py | Provides supervised embedding extraction from TabPFN models with automatic classification/regression detection and SVD-based dimensionality reduction |
| models/unsupervised/tabpfn_outlier.py | Implements unsupervised outlier model with surrogate RF for feature selection, density-aware sampling, and score calibration for probabilistic interpretation |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| finals = None | ||
| if full > final_output.shape[0]: | ||
| finals = np.full((full, 2 if self.return_flag else 1,), fill_value=0.0, dtype=np.float32) |
There was a problem hiding this comment.
The variable name 'finals' is unclear and doesn't convey its purpose. A more descriptive name like 'full_output' or 'padded_output' would better indicate that this array holds the full dataset with zero-filled values for unsampled rows.
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 2 out of 2 changed files in this pull request and generated 20 comments.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
|
Testing results:
|
|
|
1 participant
https://github.com/h2oai/h2oai/issues/34827
Due to the complexity of TabPFN, automation regression is skipped, manual testing results show down below