sweep.yaml

program: hpoptim/train_sweep_qualitai.py
method: bayes
metric:
  name: bbox_mAP_50/train
  goal: maximize
parameters:
  r_val:
    distribution: int_uniform
    min: 50
    max: 1000
  max_energy:
    distribution: int_uniform
    min: 10
    max: 100
  loss_cls_weight:
    distribution: uniform
    min: 0.05
    max: 5
  loss_bbox_weight:
    distribution: uniform
    min: 0.05
    max: 5
  loss_energy_weight:
    distribution: uniform
    min: 0.05
    max: 5
  loss_energy_gamma:
    distribution: uniform
    min: 0.2
    max: 10
  loss_energy_alpha:
    distribution: uniform
    min: 0.0001
    max: 0.9999
  lr_warmup:
    distribution: categorical
    values:
      - constant
      - linear
      - exp
  lr_warmup_iters:
    distribution: int_uniform
    min: 250
    max: 1000
  lr_warmup_ratio:
    distribution: uniform
    min: 0.16666666666666666
    max: 1
  lr:
    distribution: uniform
    min: 0.00001
    max: 0.1
early_terminate:
  type: hyperband
  eta: 35
  min_iter: 10