Subclass hydra configurations

config/callbacks/checkpoint/README.md

@@ -0,0 +1 @@
+  https://pytorch-lightning.readthedocs.io/en/stable/api/pytorch_lightning.callbacks.ModelCheckpoint.html

config/callbacks/early_stop/README.md

@@ -0,0 +1 @@
+https://pytorch-lightning.readthedocs.io/en/stable/common/early_stopping.html

config/callbacks/lr_monitor/default.yaml

@@ -0,0 +1,4 @@
+# https://pytorch-lightning.readthedocs.io/en/stable/api/pytorch_lightning.callbacks.LearningRateMonitor.html
+_target_: pytorch_lightning.callbacks.LearningRateMonitor
+logging_interval: step
+log_momentum: False

config/callbacks/richsummary/default.yaml

@@ -0,0 +1 @@
+_target_: pytorch_lightning.callbacks.RichModelSummary

config/dataloader/default.yaml

@@ -0,0 +1,5 @@
+_target_: torch.utils.data.DataLoader
+batch_size: 4
+num_workers: 2
+pin_memory: True
+dataset: ${dataset}

config/logger/wandb.yaml

@@ -0,0 +1,4 @@
+_target_: pytorch_lightning.loggers.WandbLogger
+project: brain-age
+entity: 1pha
+name: ${model.name}

config/logger/wandb_mask.yaml

@@ -0,0 +1,7 @@
+_target_: pytorch_lightning.loggers.WandbLogger
+project: brain-age
+entity: 1pha
+name: M ${dataset.mask_idx} | ${misc.seed}
+tags:
+  - model=${model.name}
+  - MASK

config/metrics/binary.yaml

@@ -0,0 +1,11 @@
+acc:
+  _target_: torchmetrics.Accuracy
+  task: binary
+  average: weighted
+f1:
+  _target_: torchmetrics.F1Score
+  task: binary
+  average: macro
+auroc:
+  _target_: torchmetrics.AUROC
+  task: binary

config/metrics/cls.yaml

@@ -0,0 +1,14 @@
+acc:
+  _target_: torchmetrics.Accuracy
+  task: multiclass
+  num_classes: ${model.backbone.num_classes}
+  average: weighted
+f1:
+  _target_: torchmetrics.F1Score
+  task: multiclass
+  num_classes: ${model.backbone.num_classes}
+  average: macro
+auroc:
+  _target_: torchmetrics.AUROC
+  task: multiclass
+  num_classes: ${model.backbone.num_classes}

config/metrics/reg.yaml

@@ -0,0 +1,9 @@
+mae:
+  _target_: torchmetrics.MeanAbsoluteError
+rmse:
+  _target_: torchmetrics.MeanSquaredError
+  squared: False
+corr:
+  _target_: torchmetrics.PearsonCorrCoef
+r2:
+  _target_: torchmetrics.R2Score

config/misc/default.yaml

@@ -0,0 +1,3 @@
+seed: 42
+debug: False
+modes: [ train, valid ]

config/module/default.yaml

@@ -0,0 +1,9 @@
+_target_: sage.trainer.PLModule
+_recursive_: False
+augmentation:
+  spatial_size: [ 160 , 192 , 160 ]
+load_from_checkpoint: # for lightning
+load_model_ckpt: # For model checkpoint only
+log_train_metrics: True
+separate_lr:
+save_dir: ${callbacks.checkpoint.dirpath}

config/train.yaml

@@ -10,104 +10,18 @@ defaults:
   - dataset: biobank
   - scheduler: cosine_anneal_warmup
   - optim: lion
-
-dataloader:
-  _target_: torch.utils.data.DataLoader
-  batch_size: 32
-  num_workers: 8
-  pin_memory: True
-  dataset: ${dataset}
-
-misc:
-  seed: 42
-  debug: False
-  modes: [ train, valid, test ]
-
-module:
-  _target_: sage.trainer.PLModule
-  _recursive_: False
-  augmentation:
-    spatial_size: [ 160 , 192 , 160 ]
-  load_from_checkpoint: # for lightning
-  load_model_ckpt: # For model checkpoint only
-  separate_lr:
-  save_dir: ${callbacks.checkpoint.dirpath}
-
-metrics:
-  mae:
-    _target_: torchmetrics.MeanAbsoluteError
-  rmse:
-    _target_: torchmetrics.MeanSquaredError
-    squared: False
-  # corr:
-  #   _target_: torchmetrics.PearsonCorrCoef
-  r2:
-    _target_: torchmetrics.R2Score
-
-logger:
-  _target_: pytorch_lightning.loggers.WandbLogger
-  project: brain-age
-  entity: 1pha
-  name: R ${model.name} | ${misc.seed}
-  tags:
-    - model=${model.name}
-    - REG
-
-# https://pytorch-lightning.readthedocs.io/en/stable/common/trainer.html
-trainer:
-  _target_: pytorch_lightning.Trainer
-  max_epochs: 500
-  devices: 1
-  accelerator: gpu
-  gradient_clip_val: 1
-  log_every_n_steps: 50
-  accumulate_grad_batches: 1
-  # DEBUGGING FLAGS. TODO: Split
-  # limit_train_batches: 0.001
-  # limit_val_batches: 0.01
+  - trainer: default
+  - dataloader: default
+  - metrics: reg
+  - logger: wandb
+  - module: default
+  - callbacks/checkpoint: reg
+  - callbacks/early_stop: reg
+  - callbacks/lr_monitor: default
+  - callbacks/richsummary: default
+  - misc: default
 
 callbacks:
-  # https://pytorch-lightning.readthedocs.io/en/stable/api/pytorch_lightning.callbacks.ModelCheckpoint.html
-  checkpoint:
-    _target_: pytorch_lightning.callbacks.ModelCheckpoint
-    dirpath: ${hydra:run.dir}
-    filename: "{step}-valid_mae{epoch/valid_MeanAbsoluteError:.3f}"
-    monitor: epoch/valid_MeanAbsoluteError
-    mode: min
-    save_top_k: 1
-    save_last: True
-    # Is useful to set it to False when metric names contain / as this will result in extra folders
-    auto_insert_metric_name: False
-
-  # checkpoint_lr:
-  #   _target_: pytorch_lightning.callbacks.ModelCheckpoint
-  #   dirpath: ${hydra:run.dir}
-  #   filename: "ckpt-step{step}-lr{_lr:.2e}-valid_mae{epoch/valid_MeanAbsoluteError:.3f}"
-  #   monitor: _lr
-  #   save_top_k: 7
-  #   mode: min
-  #   auto_insert_metric_name: False
-
-  # https://pytorch-lightning.readthedocs.io/en/stable/common/early_stopping.html
-  early_stop:
-    _target_: pytorch_lightning.callbacks.early_stopping.EarlyStopping
-    monitor: epoch/valid_MeanSquaredError
-    mode: min
-    min_delta: 0.005
-    patience: 20
-
-  # https://pytorch-lightning.readthedocs.io/en/stable/api/pytorch_lightning.callbacks.LearningRateMonitor.html
-  lr_monitor:
-    _target_: pytorch_lightning.callbacks.LearningRateMonitor
-    logging_interval: step
-    log_momentum: False
-
-  richsummary:
-    _target_: pytorch_lightning.callbacks.RichModelSummary
-
-  richpbar:
-    _target_: pytorch_lightning.callbacks.RichProgressBar
-
   manual_ckpt:
     _target_: sage.trainer.callbacks.AnchorCheckpoint
     multiplier: 1

config/train_binary.yaml

@@ -12,90 +12,13 @@ defaults:
   - dataset: ppmi
   - scheduler: exp_decay
   - optim: adamw
-
-dataloader:
-  _target_: torch.utils.data.DataLoader
-  batch_size: 4
-  num_workers: 2
-  pin_memory: True
-  dataset: ${dataset}
-
-misc:
-  seed: 42
-  debug: False
-  modes: [ train, valid ]
-
-module:
-  _target_: sage.trainer.PLModule
-  _recursive_: False
-  augmentation:
-    spatial_size: [ 160 , 192 , 160 ]
-  load_from_checkpoint: # for lightning
-  load_model_ckpt: # For model checkpoint only
-  log_train_metrics: True
-  separate_lr:
-  save_dir: ${callbacks.checkpoint.dirpath}
-
-metrics:
-  acc:
-    _target_: torchmetrics.Accuracy
-    task: binary
-    average: weighted
-  f1:
-    _target_: torchmetrics.F1Score
-    task: binary
-    average: macro
-  auroc:
-    _target_: torchmetrics.AUROC
-    task: binary
-
-logger:
-  _target_: pytorch_lightning.loggers.WandbLogger
-  project: brain-age
-  entity: 1pha
-  name: C ${model.name}
-
-# https://pytorch-lightning.readthedocs.io/en/stable/common/trainer.html
-trainer:
-  _target_: pytorch_lightning.Trainer
-  max_epochs: 300
-  devices: 1
-  accelerator: gpu
-  gradient_clip_val: 1
-  log_every_n_steps: 50
-  accumulate_grad_batches: 4
-  # DEBUGGING FLAGS. TODO: Split
-  # limit_train_batches: 0.001
-  # limit_val_batches: 0.01
-
-callbacks:
-  # https://pytorch-lightning.readthedocs.io/en/stable/api/pytorch_lightning.callbacks.ModelCheckpoint.html
-  checkpoint:
-    _target_: pytorch_lightning.callbacks.ModelCheckpoint
-    dirpath: ${hydra.run.dir}
-    filename: "{step}-valid_f1-{epoch/valid_BinaryF1Score:.3f}"
-    monitor: epoch/valid_BinaryF1Score
-    mode: max
-    save_top_k: 1
-    save_last: True
-    # Is useful to set it to False when metric names contain / as this will result in extra folders
-    auto_insert_metric_name: False
-
-  # https://pytorch-lightning.readthedocs.io/en/stable/common/early_stopping.html
-  early_stop:
-    _target_: pytorch_lightning.callbacks.early_stopping.EarlyStopping
-    monitor: epoch/valid_BinaryF1Score
-    mode: max
-    patience: 10
-
-  # https://pytorch-lightning.readthedocs.io/en/stable/api/pytorch_lightning.callbacks.LearningRateMonitor.html
-  lr_monitor:
-    _target_: pytorch_lightning.callbacks.LearningRateMonitor
-    logging_interval: step
-    log_momentum: False
-
-  richsummary:
-    _target_: pytorch_lightning.callbacks.RichModelSummary
-
-  richpbar:
-    _target_: pytorch_lightning.callbacks.RichProgressBar
+  - trainer: default
+  - dataloader: default
+  - metrics: binary
+  - logger: wandb
+  - module: default
+  - callbacks/checkpoint: binary
+  - callbacks/early_stop: binary
+  - callbacks/lr_monitor: default
+  - callbacks/richsummary: default
+  - misc: default