byol

byol/config.yaml

@@ -0,0 +1,339 @@
+# pytorch_lightning==1.8.0
+seed_everything: 123
+trainer:
+  logger:
+  - class_path: pytorch_lightning.loggers.TensorBoardLogger
+    init_args:
+      save_dir: lightning_logs
+      name: tmp
+      version: null
+      log_graph: false
+      default_hp_metric: true
+      prefix: ''
+      sub_dir: null
+      comment: ''
+      purge_step: null
+      max_queue: 10
+      flush_secs: 120
+      filename_suffix: ''
+  - class_path: pytorch_lightning.loggers.WandbLogger
+    init_args:
+      name: null
+      save_dir: .
+      version: null
+      offline: false
+      dir: null
+      id: null
+      anonymous: null
+      project: tmp
+      log_model: false
+      experiment: null
+      prefix: ''
+      job_type: null
+      config: null
+      entity: null
+      reinit: null
+      tags: null
+      group: null
+      notes: null
+      magic: null
+      config_exclude_keys: null
+      config_include_keys: null
+      mode: null
+      allow_val_change: null
+      resume: null
+      force: null
+      tensorboard: null
+      sync_tensorboard: null
+      monitor_gym: null
+      save_code: null
+      settings: null
+  enable_checkpointing: true
+  callbacks:
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-val-loss-{epoch}-{step}
+      monitor: loss/loss/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-eer-loss-{epoch}-{step}
+      monitor: EER evaluation proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-invar-val-{epoch}-{step}
+      monitor: loss/invariance/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: best-order-val-{epoch}-{step}
+      monitor: Order evaluation mean proj/val
+      verbose: false
+      save_last: null
+      save_top_k: 1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: null
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.ModelCheckpoint
+    init_args:
+      dirpath: null
+      filename: cptk-{epoch}-{step}
+      monitor: null
+      verbose: false
+      save_last: null
+      save_top_k: -1
+      save_weights_only: false
+      mode: min
+      auto_insert_metric_name: true
+      every_n_train_steps: null
+      train_time_interval: null
+      every_n_epochs: 50
+      save_on_train_epoch_end: null
+  - class_path: pytorch_lightning.callbacks.RichProgressBar
+    init_args:
+      refresh_rate: 1
+      leave: false
+      theme:
+        description: white
+        progress_bar: '#6206E0'
+        progress_bar_finished: '#6206E0'
+        progress_bar_pulse: '#6206E0'
+        batch_progress: white
+        time: grey54
+        processing_speed: grey70
+        metrics: white
+      console_kwargs: null
+  - class_path: callbacks.hypersphere.HypersphereEvaluation
+    init_args:
+      normalize: true
+      use_projections: false
+  - class_path: callbacks.hypersphere.HypersphereEvaluation
+    init_args:
+      normalize: true
+      use_projections: true
+  - class_path: callbacks.lr_logger.LearningRateLogger
+  - class_path: bernardo.callbacks.evaluation.OrderEvaluation
+    init_args:
+      log_n_epochs: 5
+      on_train: true
+      use_projection: true
+  - class_path: bernardo.callbacks.evaluation.EEREvaluation
+    init_args:
+      use_more_neg: false
+      log_n_epochs: 5
+      on_train: false
+      use_projection: true
+  default_root_dir: null
+  gradient_clip_val: null
+  gradient_clip_algorithm: null
+  num_nodes: 1
+  num_processes: null
+  devices:
+  - 1
+  gpus: null
+  auto_select_gpus: false
+  tpu_cores: null
+  ipus: null
+  enable_progress_bar: true
+  overfit_batches: 0.0
+  track_grad_norm: -1
+  check_val_every_n_epoch: 1
+  fast_dev_run: false
+  accumulate_grad_batches: null
+  max_epochs: 500
+  min_epochs: null
+  max_steps: -1
+  min_steps: null
+  max_time: null
+  limit_train_batches: null
+  limit_val_batches: null
+  limit_test_batches: null
+  limit_predict_batches: null
+  val_check_interval: null
+  log_every_n_steps: 50
+  accelerator: gpu
+  strategy: null
+  sync_batchnorm: false
+  precision: 32
+  enable_model_summary: true
+  num_sanity_val_steps: 2
+  resume_from_checkpoint: null
+  profiler: null
+  benchmark: null
+  deterministic: null
+  reload_dataloaders_every_n_epochs: 0
+  auto_lr_find: false
+  replace_sampler_ddp: true
+  detect_anomaly: false
+  auto_scale_batch_size: false
+  plugins: null
+  amp_backend: native
+  amp_level: null
+  move_metrics_to_cpu: false
+  multiple_trainloader_mode: max_size_cycle
+  inference_mode: true
+ckpt_path: null
+model:
+  class_path: models.byol.BYOL
+  init_args:
+    module:
+      class_path: networks.my_siamese_arm.SiameseArm
+      init_args:
+        encoder:
+          class_path: bernardo.models.model.Encoder
+          init_args:
+            backbone: efficientnet_b0
+            feature_extractor:
+              class_path: bernardo.models.model.FeatureExtractor
+              init_args:
+                spec_layer: melspectogram
+                n_fft: 2048
+                hop_length: 512
+            embedding_dim: 1000
+            pretrained: true
+            progress: true
+            stochastic_depth_prob: 0.2
+            norm_layer: null
+        projector:
+          class_path: bernardo.models.model.Projection
+          init_args:
+            input_dim: 1000
+            output_dim: 128
+            nonlinearity: null
+            is_identity: false
+            l2_normalize: true
+        predictor:
+          class_path: networks.mlp.MLP
+          init_args:
+            dims:
+            - 128
+            - 1024
+            - 128
+            activation: true
+            use_batchnorm: true
+            batchnorm_fn: null
+            last_layer: null
+            bias: null
+            layer_init: null
+        normalize_representations: false
+        normalize_projections: true
+    loss_fn:
+      class_path: torch.nn.MSELoss
+      init_args:
+        size_average: null
+        reduce: null
+        reduction: mean
+    weight_callback:
+      class_path: callbacks.ma_updates.MAWeightUpdate
+      init_args:
+        initial_tau: 0.99
+        max_epochs: 1000
+        should_update: true
+    optimizer:
+      class_path: utils.optim.Adam
+      init_args:
+        lr: 3.0e-05
+        betas:
+        - 0.9
+        - 0.999
+        eps: 1.0e-08
+        weight_decay: 1.5e-06
+        amsgrad: false
+    scheduler:
+      class_path: utils.optim.LinearWarmupCosineAnnealing
+      init_args:
+        warmup_epochs: 10
+        max_epochs: 1000
+        warmup_start_lr: 0.0
+        eta_min: 0.0
+        last_epoch: -1
+data:
+  class_path: bernardo.data.vocals.VocalsDataModule
+  init_args:
+    augs_neg:
+      enable: false
+      gaussian_noise: 0.5
+      pitch_shift_naive: 0
+      time_stretch: 0
+      gain: 0.5
+      shift: 0
+      parametric_eq: 0
+      tanh_distortion: 0
+      time_mask: 0
+      formant_shift_parselmouth: 0
+      pitch_shift_parselmouth: 0
+      pitch_range_parselmouth: 0
+      pitch_shift_parselmouth_prob: 0
+    positive_examples: same_clip
+    dataset_dirs:
+    - tencys_vocals
+    - ghero_vocals_3
+    - ghero_vocals_4
+    batch_size: 120
+    batch_size_val: 120
+    nr_samples: 176000
+    normalize: true
+    num_workers: 40
+    sr: 44100
+    batch_sampling_mode: sample_clips
+    eval_frac: 0.105
+    group_name_is_folder: true
+    group_by_artist: true
+    augs:
+      enable: true
+      gaussian_noise: 0.5
+      pitch_shift_naive: 0
+      time_stretch: 0
+      gain: 0.5
+      shift: 0
+      parametric_eq: 0
+      tanh_distortion: 0
+      time_mask: 0.5
+      formant_shift_parselmouth: 0
+      pitch_shift_parselmouth:
+      - 1
+      - 1.4
+      pitch_range_parselmouth: 1.5
+      pitch_shift_parselmouth_prob: 0.5
+    transform_override: false
+    verbose: true
+    use_random_loader: false
+    max_groups: -1
+    multi_epoch: 1
+    classification: false

byol/hyperparams.yaml

@@ -0,0 +1,8 @@
+encoder:
+  backbone: efficientnet_b0
+  embedding_dim: 1000
+  pretrained: false
+feature_extractor:
+  hop_length: 512
+  n_fft: 2048
+  spec_layer: melspectogram

byol/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:509f96f23a1076325f4164fa4176fa282e709da203d30b109fa05b1b6cb3c978
+size 38765343