Training

Prerequisites

uv sync --all-groups
bash scripts/sync_data.sh /path/to/kaggle/chest_xray
uv run python -m chest_xray_classifier.data.prepare --raw data/raw --out data/processed

Main — ConvNeXt-V2-Tiny

uv run python -m chest_xray_classifier.training.train experiment=sota

Expected wall time: ~90 min on an A10/A100. Checkpoint written to artifacts/checkpoints/best.ckpt.

Baseline — DINOv2 linear probe

uv run python -m chest_xray_classifier.training.train \
  model=baseline \
  trainer.max_epochs=20 \
  trainer.output_dir=artifacts/baseline

MLflow tracking

mlflow ui --backend-store-uri ./mlruns

Browse at http://localhost:5000 — every Hydra run is one MLflow run with the full resolved config logged as params and train/loss, val/loss, val/acc, val/f1_macro as metrics.

Hydra overrides (common)

Override	Effect
trainer.max_epochs=50	Longer training
trainer.accelerator=gpu	Force GPU
data.batch_size=64	Larger batches
model.lr=1e-4	Different learning rate
seed=7	Reproducibility

Multi-run sweep example:

uv run python -m chest_xray_classifier.training.train -m \
  model.lr=1e-5,3e-5,1e-4 \
  data.batch_size=32,64