feat: checkpoint wandb logging

feat: wandb config file

Former-commit-id: 45f56db86ca269b028cf76bf5315bc0eef8d2a21 [formerly e320b72e16eed02bdca05245e7c77914f0e288f9]
Former-commit-id: 6d91318784748308c73dc6a164653f04ae46cd2a

.editorconfig

@@ -11,5 +11,5 @@ charset = utf-8
 trim_trailing_whitespace = true
 insert_final_newline = true
 
-[*.{json,toml}]
+[*.{json,toml,yaml,yml}]
 indent_size = 2

.vscode/launch.json

@@ -18,7 +18,7 @@
         "--model",
         "good.onnx",
       ],
-      "justMyCode": true
+      "justMyCode": false
     }
   ]
 }

src/config-defaults.yaml

@@ -0,0 +1,41 @@
+DIR_TRAIN_IMG:
+  value: "/home/lilian/data_disk/lfainsin/train/"
+DIR_VALID_IMG:
+  value: "/home/lilian/data_disk/lfainsin/test_batched_fast/"
+DIR_SPHERE:
+  value: "/home/lilian/data_disk/lfainsin/spheres+real/"
+
+FEATURES:
+  value: { 8, 16, 32, 64 }
+N_CHANNELS:
+  value: 3,
+N_CLASSES:
+  value: 1,
+
+AMP:
+  value: True
+PIN_MEMORY:
+  value: True
+BENCHMARK:
+  value: True
+DEVICE:
+  value: gpu
+WORKERS:
+  value: 8
+
+IMG_SIZE:
+  value: 512
+SPHERES:
+  value: 5
+
+EPOCHS:
+  value: 10
+BATCH_SIZE:
+  value: 16
+
+LEARNING_RATE:
+  value: 1e-4
+WEIGHT_DECAY:
+  value: 1e-8
+MOMENTUM:
+  value: 0.9

src/data/dataloader.py

@@ -36,19 +36,6 @@ class Spheres(pl.LightningDataModule):
             pin_memory=wandb.config.PIN_MEMORY,
         )
 
-        # dataset = LabeledDataset(image_dir="/home/lilian/data_disk/lfainsin/prerender/")
-        # dataset = LabeledDataset(image_dir=wandb.config.DIR_VALID_IMG)
-        # dataset = Subset(dataset, list(range(0, len(dataset), len(dataset) // 100 + 1)))
-
-        # return DataLoader(
-        #     dataset,
-        #     shuffle=True,
-        #     batch_size=8,
-        #     prefetch_factor=8,
-        #     num_workers=wandb.config.WORKERS,
-        #     pin_memory=wandb.config.PIN_MEMORY,
-        # )
-
     def val_dataloader(self):
         dataset = LabeledDataset(image_dir=wandb.config.DIR_VALID_IMG)
         dataset = Subset(dataset, list(range(0, len(dataset), len(dataset) // 100 + 1)))
@@ -56,7 +43,7 @@ class Spheres(pl.LightningDataModule):
         return DataLoader(
             dataset,
             shuffle=False,
-            batch_size=1,
+            batch_size=8,
             prefetch_factor=8,
             num_workers=wandb.config.WORKERS,
             pin_memory=wandb.config.PIN_MEMORY,

src/train.py

@@ -1,44 +1,21 @@
 import logging
 
 import pytorch_lightning as pl
-import torch
+from pytorch_lightning.callbacks import RichProgressBar
-from pytorch_lightning.callbacks import ModelCheckpoint, RichProgressBar
 from pytorch_lightning.loggers import WandbLogger
 
 import wandb
 from data import Spheres
 from unet import UNetModule
-from utils import TableLog
+from utils import ArtifactLog, TableLog
-
-CONFIG = {
-    "DIR_TRAIN_IMG": "/home/lilian/data_disk/lfainsin/train/",
-    "DIR_VALID_IMG": "/home/lilian/data_disk/lfainsin/test_batched_fast/",
-    "DIR_SPHERE": "/home/lilian/data_disk/lfainsin/spheres+real/",
-    "FEATURES": [8, 16, 32, 64],
-    "N_CHANNELS": 3,
-    "N_CLASSES": 1,
-    "AMP": True,
-    "PIN_MEMORY": True,
-    "BENCHMARK": True,
-    "DEVICE": "gpu",
-    "WORKERS": 8,
-    "EPOCHS": 10,
-    "BATCH_SIZE": 16,
-    "LEARNING_RATE": 1e-4,
-    "WEIGHT_DECAY": 1e-8,
-    "MOMENTUM": 0.9,
-    "IMG_SIZE": 512,
-    "SPHERES": 5,
-}
 
 if __name__ == "__main__":
     # setup logging
     logging.basicConfig(level=logging.INFO, format="%(levelname)s: %(message)s")
 
-    # setup wandb
+    # setup wandb, config loaded from config-default.yaml
     logger = WandbLogger(
         project="U-Net",
-        config=CONFIG,
         settings=wandb.Settings(
             code_dir="./src/",
         ),
@@ -49,44 +26,38 @@ if __name__ == "__main__":
 
     # Create network
     model = UNetModule(
-        n_channels=CONFIG["N_CHANNELS"],
+        n_channels=wandb.config.N_CHANNELS,
-        n_classes=CONFIG["N_CLASSES"],
+        n_classes=wandb.config.N_CLASSES,
-        batch_size=CONFIG["BATCH_SIZE"],
+        batch_size=wandb.config.BATCH_SIZE,
-        learning_rate=CONFIG["LEARNING_RATE"],
+        learning_rate=wandb.config.LEARNING_RATE,
-        features=CONFIG["FEATURES"],
+        features=wandb.config.FEATURES,
     )
 
     # load checkpoint
-    state_dict = torch.load("checkpoints/synth.pth")
+    # state_dict = torch.load("checkpoints/synth.pth")
-    state_dict = dict([(f"model.{key}", value) for key, value in state_dict.items()])
+    # state_dict = dict([(f"model.{key}", value) for key, value in state_dict.items()])
-    model.load_state_dict(state_dict)
+    # model.load_state_dict(state_dict)
 
     # log gradients and weights regularly
     logger.watch(model, log="all")
 
-    # create checkpoint callback
-    checkpoint_callback = ModelCheckpoint(
-        dirpath="checkpoints",
-        filename="model.ckpt",
-        monitor="val/dice",
-    )
-
     # Create the dataloaders
     datamodule = Spheres()
 
     # Create the trainer
     trainer = pl.Trainer(
-        max_epochs=CONFIG["EPOCHS"],
+        max_epochs=wandb.config.EPOCHS,
-        accelerator=CONFIG["DEVICE"],
+        accelerator=wandb.config.DEVICE,
-        benchmark=CONFIG["BENCHMARK"],
+        benchmark=wandb.config.BENCHMARK,
         # profiler="simple",
         # precision=16,
         logger=logger,
         log_every_n_steps=1,
-        val_check_interval=25,
+        val_check_interval=100,
-        callbacks=[RichProgressBar(), checkpoint_callback, TableLog()],
+        callbacks=[RichProgressBar(), ArtifactLog(), TableLog()],
     )
 
+    # actually train the model
     trainer.fit(model=model, datamodule=datamodule)
 
     # stop wandb

src/utils/__init__.py

@@ -1,2 +1,2 @@
-from .callback import TableLog
+from .callback import ArtifactLog, TableLog
 from .paste import RandomPaste

src/utils/callback.py

@@ -1,5 +1,6 @@
+import numpy as np
+import torch
 from pytorch_lightning.callbacks import Callback
-from torch import tensor
 
 import wandb
 
@@ -22,6 +23,7 @@ class TableLog(Callback):
 
     def on_validation_batch_end(self, trainer, pl_module, outputs, batch, batch_idx, dataloader_idx):
         # unpacking
+        if batch_idx == 0:
             images, ground_truth = batch
             metrics, predictions = outputs
 
@@ -62,3 +64,30 @@ class TableLog(Callback):
                 )
             }
         )
+
+
+class ArtifactLog(Callback):
+    def on_validation_epoch_start(self, trainer, pl_module):
+        self.dices = []
+        self.best = 1
+
+    def on_validation_batch_end(self, trainer, pl_module, outputs, batch, batch_idx, dataloader_idx):
+        # unpacking
+        metrics, _ = outputs
+        self.dices.append(metrics["dice"].cpu())
+
+    def on_validation_epoch_end(self, trainer, pl_module):
+        dice = np.mean(self.dices)
+        self.dices = []
+
+        if dice < self.best:
+            self.best = dice
+
+            # create checkpoint
+            torch.save(self.state_dict(), "checkpoints/model.pth")
+            # trainer.save_checkpoint("example.ckpt") # TODO: change to .ckpt
+
+            # create and log artifact
+            artifact = wandb.Artifact("pth", type="model")
+            artifact.add_file("checkpoints/model.pth")
+            wandb.run.log_artifact(artifact)