AUPRC — APLoss + SOAP

Optimizes Average Precision (proxy for AUPRC) using the stochastic AP optimizer. Best when precision-recall tradeoff matters more than ROC.

Config

recipes/config_auprc.yaml

dataset:
  name: cifar10
  eval_splits: [val, test]
  kwargs:
    imratio: 0.02

model:
  name: resnet18
  pretrained: false
  num_classes: 1
  in_channels: 3

metrics:
  - AUPRC

training:
  project_name: libauc
  experiment_name: resnet18_APLoss_cifar10
  SEED: 2026
  epochs: 60
  batch_size: 128
  eval_batch_size: 256
  sampling_rate: 0.5
  num_workers: 0
  decay_epochs: [0.5, 0.75]
  loss: APLoss
  optimizer: SOAP
  output_path: ./output
  resume_from_checkpoint: false
  save_checkpoint_every: 5

automax:
  deterministic: true
  n_trials: 5
  SEED: 42
  name: resnet18_APLoss_cifar10
  output_directory: ./automax_output
  overwrite: true

Run

python -m src.auto_trainer \
  --config_file recipes/config_auprc.yaml