Performance logging

Summary: Pull Request resolved: #385

Test Plan: .

Differential Revision: D19739656

Pulled By: vreis

fbshipit-source-id: 347772745f2811bf2947128a23986161395c526d

classy_vision/hooks/tensorboard_plot_hook.py

@@ -140,6 +140,21 @@ def on_phase_end(
                     )
                     continue
 
+        if hasattr(task, "perf_log"):
+            for perf in task.perf_log:
+                phase_idx = perf["phase_idx"]
+                tag = perf["tag"]
+                for metric_name, metric_value in perf.items():
+                    if metric_name in ["phase_idx", "tag"]:
+                        continue
+
+                    self.tb_writer.add_scalar(
+                        f"Performance/{tag}/{metric_name}",
+                        metric_value,
+                        global_step=phase_idx,
+                    )
+            task.perf_log = []
+
         # flush so that the plots aren't lost if training crashes soon after
         self.tb_writer.flush()
         logging.info(f"Done plotting to Tensorboard")

classy_vision/tasks/classification_task.py

@@ -7,6 +7,7 @@
 import copy
 import enum
 import logging
+import time
 from typing import Any, Dict, List, Optional, Union
 
 import torch
@@ -807,18 +808,53 @@ def get_global_batchsize(self):
     def on_start(self, local_variables):
         self.run_hooks(local_variables, ClassyHookFunctions.on_start.name)
 
+        self.perf_log = []
+
     def on_phase_start(self, local_variables):
+        self.phase_start_time_total = time.perf_counter()
+
         self.advance_phase()
 
         self.run_hooks(local_variables, ClassyHookFunctions.on_phase_start.name)
 
+        self.phase_start_time_train = time.perf_counter()
+
     def on_phase_end(self, local_variables):
+        self.log_phase_end("train")
+
         logging.info("Syncing meters on phase end...")
         for meter in self.meters:
             meter.sync_state()
         logging.info("...meters synced")
         barrier()
+
         self.run_hooks(local_variables, ClassyHookFunctions.on_phase_end.name)
 
+        self.log_phase_end("total")
+
     def on_end(self, local_variables):
         self.run_hooks(local_variables, ClassyHookFunctions.on_end.name)
+
+    def log_phase_end(self, tag):
+        if not self.train:
+            return
+
+        assert self.phase_type == "train"
+
+        start_time = (
+            self.phase_start_time_train
+            if tag == "train"
+            else self.phase_start_time_total
+        )
+        phase_duration = time.perf_counter() - start_time
+        im_per_sec = (
+            self.get_global_batchsize() * len(self.dataloaders[self.phase_type])
+        ) / phase_duration
+        self.perf_log.append(
+            dict(
+                tag=tag,
+                phase_idx=self.train_phase_idx,
+                epoch_duration=phase_duration,
+                im_per_sec=im_per_sec,
+            )
+        )