seed: 3407

dataset:
  name: 'CoordinatesData'
  args:
    data_root: '/Users/zhouweiqi/Downloads/gcfp/data/dataset'
    train_anno_file: '/Users/zhouweiqi/Downloads/gcfp/data/dataset/train.csv'
    val_anno_file: '/Users/zhouweiqi/Downloads/gcfp/data/dataset/valid.csv'

dataloader:
  batch_size: 32
  num_workers: 4
  pin_memory: true
  shuffle: true

model:
  name: 'VisionTransformer'
  args:
    img_size: 224
    patch_size: 16
    in_c: 3
    num_classes: 5
    embed_dim: 8
    depth: 12
    num_heads: 2
    mlp_ratio: 4.0
    qkv_bias: true
    qk_scale: null
    representation_size: null
    distilled: false 
    drop_ratio: 0.
    attn_drop_ratio: 0.
    drop_path_ratio: 0.
    norm_layer: null
    act_layer: null
    input_length: 29

solver:
  name: 'VITSolver'
  args:
    epoch: 100
    no_other: false
    base_on: null
    model_path: null

  optimizer:
    name: 'Adam'
    args:
      lr: !!float 1e-3
      # weight_decay: !!float 5e-5

  lr_scheduler:
    name: 'CosineLR'
    args:
      epochs: 100
      lrf: 0.1

  loss:
    name: 'SigmoidFocalLoss'
    # name: 'CrossEntropyLoss'
    args:
      reduction: "mean"
      alpha: 0.95

  logger:
    log_root: '/Users/zhouweiqi/Downloads/test/logs'
    suffix: 'vit'