run_config.yaml

mp3:
  song_hz: 22050
  inst_len: 10.0
  skip_interval: 5.0

feature:
  n_bins: 144
  bins_per_octave: 24
  hop_length: 2048
  large_voca: False
#  large_voca: True

experiment:
  learning_rate : 0.0001
  weight_decay : 0.0
  max_epoch : 100
  batch_size : 128
  save_step : 40
  data_ratio : 0.8

model:
  feature_size : 144
  timestep : 108
  num_chords : 25
#  num_chords : 170
  input_dropout : 0.2
  layer_dropout : 0.2
  attention_dropout : 0.2
  relu_dropout : 0.2
  num_layers : 8
  num_heads : 4
  hidden_size : 128
  total_key_depth : 128
  total_value_depth : 128
  filter_size : 128
  loss : 'ce'
  probs_out : False

path:
  ckpt_path : 'model'
  result_path : 'result'
  asset_path : '/data/music/chord_recognition/jayg996/assets'
  root_path : '/data/music/chord_recognition'