update

Author: hughplay

conf/config.yaml

@@ -5,6 +5,8 @@ defaults:
   - scheduler: step
   - dataset: imagenet
   - pipeline: classifier
+  - hydra/job_logging: colorlog
+  - hydra/hydra_logging: colorlog
 
 pl_trainer:
   accelerator: ddp
@@ -36,6 +38,8 @@ callbacks:
     monitor: val_acc1
     save_top_k: 2
     verbose: true
+  - _target_: pytorch_lightning.callbacks.LearningRateMonitor
+    logging_interval: 'step'
 
 run_test: true
 seed: 2021

conf/optim/adam.yaml

@@ -0,0 +1,3 @@
+# @package _group_
+_target_: torch.optim.Adam
+lr: 0.001

conf/optim/adamw.yaml

@@ -0,0 +1,3 @@
+# @package _group_
+_target_: torch.optim.AdamW
+lr: 0.001

conf/scheduler/cosine.yaml

@@ -0,0 +1,3 @@
+# @package _group_
+_target_: torch.optim.lr_scheduler.CosineAnnealingLR
+T_max: 90

docker/requirements.txt

@@ -1,4 +1,5 @@
 ipython
 pytorch-lightning==1.2.5
 hydra-core==1.0.6
+hydra_colorlog
 omegaconf==2.0.6

pipeline/classifier.py

@@ -32,13 +32,6 @@ def __init__(
     def forward(self, x):
         return self.model(x)
 
-    def _log_metrics(self):
-        if self.trainer.is_global_zero:
-            str_metrics = ''
-            for key, val in self.trainer.logged_metrics.items():
-                str_metrics += f'\n\t{key}: {val}'
-            logger.info(str_metrics)
-
     def training_step(self, batch, batch_idx):
         images, target = batch
         output = self(images)
@@ -48,7 +41,7 @@ def training_step(self, batch, batch_idx):
         self.log('train_acc1', acc1, on_step=True, prog_bar=True, on_epoch=True)
         self.log('train_acc5', acc5, on_step=True, on_epoch=True)
         return loss_train
-    
+
     def validation_step(self, batch, batch_idx):
         images, target = batch
         output = self(images)
@@ -59,9 +52,6 @@ def validation_step(self, batch, batch_idx):
         self.log('val_acc5', acc5, on_epoch=True)
         return loss_val
 
-    def on_validation_end(self):
-        self._log_metrics()
-
     @staticmethod
     def __accuracy(output, target, topk=(1, )):
         """Computes the accuracy over the k top predictions for the specified values of k"""
@@ -89,6 +79,16 @@ def test_step(self, batch, batch_idx):
         self.log('test_acc5', acc5, on_epoch=True)
         return loss_test
 
+    def _log_metrics(self):
+        if self.trainer.is_global_zero:
+            str_metrics = ''
+            for key, val in self.trainer.logged_metrics.items():
+                str_metrics += f'\n\t{key}: {val}'
+            logger.info(str_metrics)
+
+    def on_validation_end(self):
+        self._log_metrics()
+
     def on_test_end(self):
         self._log_metrics()