TCN Bounty Implementation

[bilalarif3197]

Bilal Arif
netID: barif

Contribution Type: Updated model

Description:

• Updated to current BaseModel/EmbeddingModel API
• Implemented dilated causal convolutions for temporal sequence modeling on EHR data
• Supports both discrete code sequences (diagnoses, procedures) and continuous timeseries
• Includes comprehensive test suite with 6 passing unit tests
• Added example notebook demonstrating mortality prediction on MIMIC-III

Files to review:

• pyhealth/models/tcn.py - Full TCN implementation
• tests/core/test_tcn.py - Main testing suite
• examples/tcn_mimic3_codes.ipynb - Jupyter notebook for using MIMIC-III

[Copilot]

Pull request overview

This PR successfully updates the TCN (Temporal Convolutional Networks) model to the current PyHealth 2.0 API, replacing the deprecated SampleEHRDataset with SampleDataset and adopting the EmbeddingModel architecture. The implementation provides dilated causal convolutions for temporal sequence modeling on EHR data, supporting both discrete code sequences (diagnoses, procedures) and continuous timeseries data.

Key changes:

• Modernized API: Updated from legacy SampleEHRDataset to SampleDataset with EmbeddingModel
• Simplified architecture: Removed complex manual feature processing in favor of EmbeddingModel's automatic handling
• Improved PyTorch compatibility: Updated weight_norm import to PyTorch 2.0+ path (torch.nn.utils.parametrizations)

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
pyhealth/models/tcn.py	Complete TCN model refactor to PyHealth 2.0 API; removed 150+ lines of legacy code; updated weight_norm import; added safety check in Chomp1d; improved forward pass with explicit last-output extraction
tests/core/test_tcn.py	New comprehensive test suite with 6 test cases covering initialization, forward/backward passes, embeddings, and custom hyperparameters
examples/tcn_mimic3_codes.ipynb	New example notebook demonstrating mortality prediction on MIMIC-III dataset with clear step-by-step instructions

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Potential StopIteration error if self.feature_keys is empty. Consider adding a check:

if not self.feature_keys:
    raise ValueError("At least one input feature is required")

before line 240, or handle the case where self.tcn might be empty.

[LogicFan]

Could you also add usage examples in the new docstring?

[bilalarif3197]

I added a usage example in the docstring. I also changed SampleDataset to create_sample_dataset in the test file.