inference/data/base_config.yaml

version: "1.34"

lr: 1e-4
log_step: 1
split: 0
batch_size: 8
sr: 16000

datasets:
  - jamendo
  - emomusic
  - pmemo
  - deam

model:
  encoder: "MERT"
  layers:
    - 5
    - 6
  classifier: "linear-mt-attn-ck"
  # - linear
  # - linear-attn-ck
  # - linear-mt-attn-ck

  kd: True
  kd_weight: 0.8
  kd_temperature: 1
  lr: 1e-4

# audio_path: './dataset/jamendo'
# subset: 'moodtheme'

dataset:
  jamendo:
    root: './dataset/jamendo'
    subset: 'moodtheme'
    batch_size: 8
    output_size : 56
    split: 0
    segment_type: "all" # [all,f10s,f30s,10s,30s]
    num_workers: 4
  deam:
    root: './dataset/deam'
    batch_size: 8
    output_size : 2 
    segment_type: "all" # [all,f10s,f30s,10s,30s]
    num_workers: 4
  pmemo:
    root: './dataset/pmemo'
    batch_size: 8
    output_size : 2 
    segment_type: "all" # [all,f10s,f30s,10s,30s]
    num_workers: 4
  emomusic:
    root: './dataset/emomusic'
    batch_size: 8
    output_size : 2 
    segment_type: "all" # [all,f10s,f30s,10s,30s]
    num_workers: 4



# --------------------------------------- #
genre_class_size: 87
mood_class_size: 56
instr_class_size: 40
dac_latents_size: 72
dac_rvq_size: 9
# --------------------------------------- #


#PMEMO BEST (0.5360 0.7772), mt: (0.5401 0.7780)
checkpoint_pmemo: "tb_logs/best/P.ckpt"

#DEAM BEST (0.5131 0.6025), mt: (0.5150 0.6125)
checkpoint_deam: "tb_logs/best/D.ckpt"

#EMOMUSIC BEST (0.5957 0.7489), mt: (0.6091 0.7525)
checkpoint_emomusic: "tb_logs/best/E.ckpt"

#JAMENDO BEST (0.1521 0.7806)
checkpoint_jamendo: "tb_logs/best/J.ckpt"




# datasets:
#   - jamendo
#   - pmemo
#   - deam
#   - emomusic
#   - pmemo
#   - jamendo
# datasets_val:
#   - emomusic
# model_save_path: './saved_models/'
# results_save_path: './results/'
# hydra:
#   job:
#     chdir: True
# - MERT M2L LIBROSA Encodec DAC
# aggr_method: "mean"
# - mean
# - median
# - 80th_percentile
# - max