fix(training): drop removed ReduceLROnPlateau verbose arg (#22)

[bradsmithmba]

Summary

ReduceLROnPlateau was constructed with verbose=True in two places:

• src/models/transfer_trainer.py — TransferTrainer.setup_optimizers()
• src/training/trainer.py — Trainer.__init__

PyTorch deprecated the scheduler verbose parameter in 2.2 and removed it afterward. On current torch the constructor raises TypeError: ReduceLROnPlateau.__init__() got an unexpected keyword argument 'verbose', so any path that builds the scheduler fails. This broke 3 tests in tests/models/test_transfer_trainer.py at optimizer setup.

Closes #22.

Fix

Remove verbose=True from both call sites. Both sites were found (the issue mentioned transfer_trainer; trainer.py has the identical defect and is fixed here too).

Verification

On torch 2.12.0:

ReduceLROnPlateau(opt, mode='min', factor=0.5, patience=5)              -> constructs + steps OK
ReduceLROnPlateau(opt, mode='min', factor=0.5, patience=10, min_lr=1e-6) -> constructs + steps OK
ReduceLROnPlateau(opt, ..., verbose=True)                              -> TypeError (old behavior)

CI note

tests/models/test_transfer_trainer.py transitively imports the src.features chain, which is broken until #1 (PR #19) merges, so the file can't be collected on master yet. The corrected scheduler construction was verified directly against torch 2.12. The 3 tests run once #1 lands.

🤖 Generated with Claude Code