config.yaml

# path:
#   Path to dataset which subdirectories corresponding to target labels (e.g. positive/negative).
# training/validation_pct:
#   % of training/validation subsets after train_test_split.
#   Test set % will be calculated as (1 - training_pct - validation_pct).
dataset:
  path: "D:/GitHub/datasets/crack-detection"
  # path: "D:/GitHub/datasets/SDNET2018/pavements"
  training_pct: 0.8
  validation_pct: 0.1

# Configuration for Neptune logger.
# offline_mode:
#   Set to True will send log data to Neptune, turn it off when only debugging.
neptune:
  offline_mode: False
  project_name: 'mtszkw/surface-crack-detect'
  experiment_name: 'Transfer-learning (AlexNet)'

# Optimizer parameters
optimizer:
  lr: 0.0005
  amsgrad: False

# Scheduler parameters
scheduler:
  step_size: 1
  gamma: 0.8

# Training
# use_gpu:
#   Passed to pytorch_lightning.Trainer as number of GPUs to be used for training.
# max_epochs:
#   Maximum number of training epochs
# batch_size:
#   Batch size
# val_check_interval:
#   Contols how often validation set should be used (if float = % epochs, if int = # steps).
training:
  use_gpu: 1
  max_epochs: 5
  batch_size: 64
  val_check_interval: 0.2

# Debug features
# train/val/test_percent_check:
#   Used in pytorch_lightning.Trainer.fit method, allows to use only a % of training/validation/test set.
#   Useful for quick runs.
debugging:
  train_percent_check: 1.0
  val_percent_check: 1.0
  test_percent_check: 1.0