PR from PyHealth_for_labrador#1048
Open
Haoming0161 wants to merge 44 commits intosunlabuiuc:masterfrom
Open
Conversation
…-implementation Improve Labrador model tests with synthetic fixtures
Improve Labrador model tests with synthetic fixtures
…-implementation-tc2cpe Add MLM head and losses to LabradorModel, make classifier optional, and expand tests
…-implementation-jyoaeb Labrador: add joint value embedding, MLM head, losses, model refactor and tests
Align Labrador tests & model
…-implementation-drlpou Motivation This PR contributes a PyHealth adaptation of the Labrador architecture for lab-centric modeling. The goal is to provide a review-friendly model contribution with complete required components: model implementation, API docs, index registration, ablation example, and fast synthetic tests. Paper reference: Bellamy et al., Labrador: Exploring the Limits of Masked Language Modeling for Laboratory Data (ML4H 2024): https://arxiv.org/abs/2312.11502. Required Files (Course Checklist) 1) Core implementation (model) pyhealth/models/labrador.py: implements LabradorModel, LabradorEmbedding, LabradorValueEmbedding, and LabradorMLMHead with optional classifier/MLM heads and masking-aware losses. 2) API docs docs/api/models/pyhealth.models.Labrador.rst: model doc page with paper link, main model docs, supporting modules, and usage pointer. 3) Index update docs/api/models.rst: added Labrador to the models toctree for discoverability. 4) Example / ablation script examples/mimic4_mortality_labrador.py: synthetic fast ablation-style example varying hidden_dim, num_layers, and dropout. 5) Tests tests/test_labrador.py: synthetic unit tests for embedding outputs, validation errors, MLM head outputs/losses, forward behavior, all-padding stability, and gradient flow. Example Run the ablation-style example: python examples/mimic4_mortality_labrador.py What it does: Builds synthetic lab sequences and binary labels. Splits train/validation sets. Trains/evaluates multiple Labrador hyperparameter configs quickly. Tests Executed commands: pytest -q tests/test_labrador.py --durations=10 python -m py_compile pyhealth/models/labrador.py examples/mimic4_mortality_labrador.py Observed: tests/test_labrador.py: all tests passed. Per-case timings are in milliseconds range (fast unit-test style). Compile checks passed. Known Limits The example script is intentionally synthetic for speed and reproducibility; it is not intended as a full real-dataset benchmark. Total pytest wall-clock time includes Python/pytest startup and module-import overhead; per-testcase execution remains very fast.
…-implementation-dr4v5o Implement Labrador model with MLM/value heads, docs, example, and tests
Polish Labrador model docs and add structured ablation example
…-for-model-implementation-ghfhpt Add Labrador model (embeddings, MLM head), docs, example, and tests
PR description updated
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Motivation
Reproduce and extend the Labrador paper in PyHealth as a Model contribution (Option 2).
Provide a clear, review-friendly implementation with complete docs, index registration, runnable ablation example, and fast synthetic tests.
Ensure contribution quality aligns with course rubric (implementation, documentation, code style, testing, and PR formatting).
Contributors (Group)
Ying Liang (ying24)
Haoming Qin (hqin11)
Yuhan Ding (yuhand7)
Type of Contribution
Option 2: Model contribution
Original Paper
Bellamy et al., Labrador: Exploring the Limits of Masked Language Modeling for Laboratory Data (ML4H 2024)
Link: https://arxiv.org/abs/2312.11502
High-level Description
Added LabradorModel (inherits BaseModel) with:
joint lab-code/value embedding
transformer encoder
optional classifier head
optional MLM head for categorical + continuous predictions
masking-aware MLM losses (categorical_mlm_loss, continuous_mlm_loss)
Added module-level and class-level documentation with usage context and paper citation.
Added compact ablation example using synthetic data and hyperparameter variations.
Added/updated fast synthetic unit tests for model instantiation, forward behavior, output shapes, loss computation, padding edge case, and gradient flow.
File Guide (What to Review)
Core model
pyhealth/models/labrador.py
Model API docs
docs/api/models/pyhealth.models.Labrador.rst
Docs index update
docs/api/models.rst
Example / ablation
examples/mimic4_mortality_labrador.py
Tests
tests/test_labrador.py
Testing
pytest -q tests/test_labrador.py --durations=10
python -m py_compile pyhealth/models/labrador.py tests/test_labrador.py examples/mimic4_mortality_labrador.py
Result: all Labrador tests pass; per-test runtime remains very small (milliseconds-level for test-call time).
Known Limits
Example ablation uses synthetic data for fast/reproducible CI-style execution; it is a demonstration script, not a full real-dataset benchmark pipeline.
Total pytest wall-clock includes startup/import overhead; individual test calls remain fast.