moco_sweep.yaml

command:
- ${env}
- ${interpreter}
- ${program}
- "momentum_sweep"
- "--n_epochs=10"
- "--patience=5"
- "--tasks=contr"
- "--modelcls=models.MomentumContrastiveModel"
- "--batch_size=1"
- "--K=50000"
- ${args}
program: train.py
early_terminate:
  min_iter: 3
  type: hyperband
method: bayes
metric:
  goal: maximise
  name: val/earlystop_criterium
parameters:
  batch_update:
    distribution: int_uniform
    max: 32
    min: 16
  dem_dropout:
    distribution: uniform
    max: 0.6
    min: 0.15
  event_dropout:
    distribution: uniform
    max: 0.3
    min: 0.075
  freeze_emb:
    distribution: categorical
    values:
    - "True"
    - "False"
  hidden_size:
    distribution: int_uniform
    max: 100
    min: 50
  lr:
    distribution: uniform
    max: 0.01
    min: 0.0001
  min_word_count:
    distribution: int_uniform
    max: 1000
    min: 5
  momentum:
    distribution: uniform
    max: 0.999
    min: 0.99
  normalize:
    distribution: categorical
    values:
    - models.Normalizer
    - models.LNNormalizer
  normalize_mlp:
    distribution: categorical
    values:
    - "True"
    - "False"
  padaware:
    distribution: categorical
    values:
    - "True"
    - "False"
  patient_modelcls:
    distribution: categorical
    values:
    - models.PatientPoolEncoder
    - models.PatientRNNEncoder
  prediction_steps:
    distribution: int_uniform
    max: 4
    min: 1
  rand_emb:
    distribution: categorical
    values:
    - "True"
    - "False"
  sampler:
    distribution: categorical
    values:
    - samplers.DiagnoseAgeSubjectRandomSampler
    - torch.utils.data.RandomSampler
  step_dropout:
    distribution: uniform
    max: 0.2
    min: 0.05
  temperature:
    distribution: uniform
    max: 0.5
    min: 0.07
  timestep_modelcls:
    distribution: categorical
    values:
    - models.LinearMaxMeanSumPool
    - models.LinearMaxSumPool
    - models.LinearMaxPool