Remove dead pre-2.0 VAE and Graph_TorchvisionModel - #1210
Conversation
Both call BaseModel.__init__ with the pre-2.0 (dataset, feature_keys, label_key, mode) signature and always raise TypeError on construction; they also use other removed pre-2.0 APIs and are referenced nowhere in the library or tests. Remove the modules, their exports, and the dangling VAE autodoc page.
|
These removals leave two tracked examples importing names that no longer exist: |
done |
2 participants
Issue
VAE (vae.py) and Graph_TorchvisionModel (graph_torchvision_model.py) call BaseModel.__init__ with the pre-2.0 signature (dataset, feature_keys, label_key, mode), but the current BaseModel.__init__ accepts only dataset. Instantiating either class therefore always raises TypeError, so both are unusable. They are pre-2.0 dead code: beyond the constructor, Graph_TorchvisionModel also calls get_label_tokenizer(), get_output_size(tokenizer), and prepare_labels(), none of which exist on the current BaseModel. Both classes are exported from pyhealth.models but referenced nowhere in the library or test suite.
Fix
Removed the two dead modules (pyhealth/models/vae.py, pyhealth/models/graph_torchvision_model.py) and their exports in pyhealth/models/__init__.py. Also removed the now-dangling VAE autodoc page (docs/api/models/pyhealth.models.VAE.rst) and its toctree entry in docs/api/models.rst, which would otherwise break the docs build.
Notes
Two example files still reference the removed classes: examples/cxr/chextXray_image_generation_VAE.py and examples/graph_torchvision_model.ipynb. They were already non-functional (their models could never be constructed) and were left in place to keep this pr scoped to the dead library code. Also historical mentions in docs/log.rst (changelog) and docs/tutorials.rst (example listing) werEeleft untouched as they do not break the build.