Add Labrador Model (ICML 2024) – PyTorch Port With Tests

[Rad099]

This PR contributes a new ML model to PyHealth: Labrador, a PyTorch reimplementation of the ICML 2024 model from:

Bellamy, Kumar, Wang, & Beam (2024). "Labrador: Exploring the Limits of Masked Language Modeling for Laboratory Data." ICML 2024.

Model Summary

• Transformer-based encoder for laboratory data
• Dual embedding pathways:
• categorical lab codes
• continuous lab values (with mask/null handling)
• Stacked Transformer encoder blocks
• Masked mean pooling over the sequence
• Optional static feature integration
• Downstream classifier head supporting: binary, multiclass, multilabel, regression
• Fully compatible with PyHealth BaseModel and Trainer APIs

Included in this PR

• pyhealth/models/labrador.py — full PyTorch implementation of the model
• Integration in pyhealth/models/init.py
• Unit tests (tests/core/test_labrador.py) verifying:

•   - initialization
  

•   - forward pass
  

•   - backward pass
  

•   - custom hyperparameters   
  

• Complete docstring with citation and architectural explanation

Notes

• This PR focuses on the supervised downstream variant (classification/regression).
• MLM pretraining heads from the original TensorFlow implementation are not included.