MLproject

name: notes_generator
conda_env: conda.yaml
entry_points:
  fetch:
    parameters:
      data_dir: { type: string, default: data }
    command: >
      $SHELL scripts/fetch.sh {data_dir}

  preprocess:
    parameters:
      data_dir: { type: path, default: data }
    command: >
      $SHELL scripts/preprocess.sh {data_dir}

  train:
    parameters:
      app_name: { type: string, default: STEPMANIA }
      model_dir: { type: path, default: data/onsets_model }
      score_dir: { type: path, default: data/train_data/score_onsets_1 }
      mel_dir: { type: path, default: data/train_data/mel_log }
      resume: { type: float, default: 0 }
      epochs: { type: float, default: 35 }
      batch: { type: float, default: 32 }
      lr_start: { type: float, default: 9e-4 }
      lr_end: { type: float, default: 9e-4 }
      lr_scheduler: { type: string, default: CosineAnnealingLR}
      seq_length: { type: float, default: 20480 }
      aug_count: { type: float, default: 0 }
      num_layers: { type: float, default: 2 }
      onset_weight: { type: float, default: 64 }
      dropout: { type: float, default: 0.3 }
      fuzzy_width: { type: float, default: 5 }
      fuzzy_scale: { type: float, default: 0.2 }
      with_beats: { type: float, default: 1 }
      difficulties: { type: string, default: "" }
      send_model: {type: float, default: 0}
      n_saved_model: {type: float, default: 20}
      log_artifacts: { type: string, default: "" }
      augmentation_setting: { type: string, default: loader_aug_config.yaml }
      warmup_steps: {type: float, default: 400}
      weight_decay: {type: float, default: 0}
      is_parallel: {type: float, default: 0}
      eta_min: {type: float, default: 1e-6}
      conv_stack_type: { type: string, default: "v7" }
      pretrained_model_path: {type: string, default: ""}
      rnn_dropout: {type: float, default: 0.1}
    command: >
      PYTHONPATH=. python scripts/onsets_train.py --app_name={app_name}
      --model_dir={model_dir} --score_dir={score_dir} --mel_dir={mel_dir}
      --resume={resume} --epochs={epochs} --batch={batch}
      --lr_start={lr_start} --lr_end={lr_end} --seq_length={seq_length}
      --aug_count={aug_count} --num_layers={num_layers}
      --onset_weight={onset_weight} --dropout={dropout}
      --fuzzy_width={fuzzy_width} --fuzzy_scale={fuzzy_scale}
      --with_beats={with_beats} --difficulties={difficulties}
      --send_model={send_model} --n_saved_model={n_saved_model}
      --log_artifacts={log_artifacts}
      --augmentation_setting={augmentation_setting}
      --lr_scheduler={lr_scheduler}
      --warmup_steps={warmup_steps}
      --weight_decay={weight_decay}
      --is_parallel={is_parallel}
      --eta_min={eta_min}
      --conv_stack_type={conv_stack_type}
      --pretrained_model_path={pretrained_model_path}
      --rnn_dropout={rnn_dropout}

  test:
    parameters:
      model_dir: { type: path, default: data/onsets_model }
      app_name: { type: string, default: STEPMANIA }
      score_dir: { type: path, default: data/train_data/score_onsets_1 }
      mel_dir: { type: path, default: data/train_data/mel_log }
      seq_length: { type: float, default: 20480 }
      batch: { type: float, default: 1 }
      num_layers: { type: float, default: 2 }
      onset_weight: { type: float, default: 64 }
      with_beats: { type: float, default: 1 }
      conv_stack_type: { type: string, default: "v7" }
      csv_save_dir: { type: path, default: data/model_test_result }
    command: >
      PYTHONPATH=. python scripts/model_test.py {model_dir}
      --app_name={app_name} --score_dir={score_dir}
      --mel_dir={mel_dir} --seq_length={seq_length}
      --batch={batch}
      --num_layers={num_layers}
      --onset_weight={onset_weight}
      --with_beats={with_beats}
      --conv_stack_type={conv_stack_type}
      --csv_save_dir={csv_save_dir}

  generate:
    parameters:
      model_path: { type: path, default: pretrained_model/model.pth }
      audio_path: { type: path }
      midi_save_path: { type: path, default: data/midi_out }
      bpm_info: { type: string }
    command: >
      PYTHONPATH=. python scripts/prediction_stepmania.py
      --onset_model_path={model_path}
      --audio_path={audio_path}
      --midi_save_path={midi_save_path}
      --bpm_info={bpm_info}