dw4.yaml

defaults:
  - training: default
  - training/optimizer: default_lr_schedule
  - flow: default
  - target: default
  - _self_
#  - override hydra/launcher: joblib

hydra:
  job:
    chdir: false

target:
  custom_samples: false # Allows using custom dataset instead of the default (biased) dataset.
  temperature: 0.1  # 0.1 or 1 are two good options

flow:
  dim: 2
  n_aug: 1
  nodes: 4

training:
  n_epoch: 100
  batch_size: 32
  plot_batch_size: 256
  eval_batch_size: 128
  seed: 0
  train_set_size: 1000
  test_set_size: 1000
  use_flow_aux_loss: false
  final_run: true # eval on test set instead of validation set.


logger:
#  list_logger: null
#  pandas_logger:
#    save_period: 1000 # how often to save the pandas dataframe as a csv
  wandb:
    name: dw4_${flow.type}
    project: fab
    entity: flow-ais-bootstrap
    tags: [dw4,ml,protea1]