Cleanup CheckpointSaver, add support for increasing or decreasing metric, switch to prec1 metric in train loop

train.py

@@ -93,6 +93,8 @@ parser.add_argument('--amp', action='store_true', default=False,
                     help='use NVIDIA amp for mixed precision training')
 parser.add_argument('--output', default='', type=str, metavar='PATH',
                     help='path to output folder (default: none, current dir)')
+parser.add_argument('--eval-metric', default='prec1', type=str, metavar='EVAL_METRIC',
+                    help='Best metric (default: "prec1"')
 parser.add_argument("--local_rank", default=0, type=int)
 
 
@@ -238,10 +240,13 @@ def main():
     if args.local_rank == 0:
         print('Scheduled epochs: ', num_epochs)
 
+    eval_metric = args.eval_metric
     saver = None
     if output_dir:
-        saver = CheckpointSaver(checkpoint_dir=output_dir)
-    best_loss = None
+        decreasing = True if eval_metric == 'loss' else False
+        saver = CheckpointSaver(checkpoint_dir=output_dir, decreasing=decreasing)
+    best_metric = None
+    best_epoch = None
     try:
         for epoch in range(start_epoch, num_epochs):
             if args.distributed:
@@ -255,15 +260,15 @@ def main():
                 model, loader_eval, validate_loss_fn, args)
 
             if lr_scheduler is not None:
-                lr_scheduler.step(epoch, eval_metrics['eval_loss'])
+                lr_scheduler.step(epoch, eval_metrics[eval_metric])
 
             update_summary(
                 epoch, train_metrics, eval_metrics, os.path.join(output_dir, 'summary.csv'),
-                write_header=best_loss is None)
+                write_header=best_metric is None)
 
             if saver is not None:
                 # save proper checkpoint with eval metric
-                best_loss = saver.save_checkpoint({
+                best_metric, best_epoch = saver.save_checkpoint({
                     'epoch': epoch + 1,
                     'arch': args.model,
                     'state_dict': model.state_dict(),
@@ -271,11 +276,12 @@ def main():
                     'args': args,
                     },
                     epoch=epoch + 1,
-                    metric=eval_metrics['eval_loss'])
+                    metric=eval_metrics[eval_metric])
 
     except KeyboardInterrupt:
         pass
-    print('*** Best loss: {0} (epoch {1})'.format(best_loss[1], best_loss[0]))
+    if best_metric is not None:
+        print('*** Best metric: {0} (epoch {1})'.format(best_metric, best_epoch))
 
 
 def train_epoch(
@@ -363,7 +369,7 @@ def train_epoch(
 
         end = time.time()
 
-    return OrderedDict([('train_loss', losses_m.avg)])
+    return OrderedDict([('loss', losses_m.avg)])
 
 
 def validate(model, loader, loss_fn, args):
@@ -418,7 +424,7 @@ def validate(model, loader, loss_fn, args):
                     batch_time=batch_time_m, loss=losses_m,
                     top1=prec1_m, top5=prec5_m))
 
-    metrics = OrderedDict([('eval_loss', losses_m.avg), ('eval_prec1', prec1_m.avg)])
+    metrics = OrderedDict([('loss', losses_m.avg), ('prec1', prec1_m.avg), ('prec5', prec5_m.avg)])
 
     return metrics
 

utils.py

@@ -6,6 +6,7 @@ import os
 import shutil
 import glob
 import csv
+import operator
 from collections import OrderedDict
 
 
@@ -16,24 +17,32 @@ class CheckpointSaver:
             recovery_prefix='recovery',
             checkpoint_dir='',
             recovery_dir='',
+            decreasing=False,
+            verbose=True,
             max_history=10):
 
-        self.checkpoint_files = []
+        # state
+        self.checkpoint_files = []  # (filename, metric) tuples in order of decreasing betterness
+        self.best_epoch = None
         self.best_metric = None
-        self.worst_metric = None
-        self.max_history = max_history
-        assert self.max_history >= 1
         self.curr_recovery_file = ''
         self.last_recovery_file = ''
+
+        # config
         self.checkpoint_dir = checkpoint_dir
         self.recovery_dir = recovery_dir
         self.save_prefix = checkpoint_prefix
         self.recovery_prefix = recovery_prefix
         self.extension = '.pth.tar'
+        self.decreasing = decreasing  # a lower metric is better if True
+        self.cmp = operator.lt if decreasing else operator.gt  # True if lhs better than rhs
+        self.verbose = verbose
+        self.max_history = max_history
+        assert self.max_history >= 1
 
     def save_checkpoint(self, state, epoch, metric=None):
-        worst_metric = self.checkpoint_files[-1] if self.checkpoint_files else None
-        if len(self.checkpoint_files) < self.max_history or metric < worst_metric[1]:
+        worst_file = self.checkpoint_files[-1] if self.checkpoint_files else None
+        if len(self.checkpoint_files) < self.max_history or self.cmp(metric, worst_file[1]):
             if len(self.checkpoint_files) >= self.max_history:
                 self._cleanup_checkpoints(1)
 
@@ -43,16 +52,21 @@ class CheckpointSaver:
                 state['metric'] = metric
             torch.save(state, save_path)
             self.checkpoint_files.append((save_path, metric))
-            self.checkpoint_files = sorted(self.checkpoint_files, key=lambda x: x[1])
-
-            print("Current checkpoints:")
-            for c in self.checkpoint_files:
-                print(c)
-
-            if metric is not None and (self.best_metric is None or metric < self.best_metric[1]):
-                self.best_metric = (epoch, metric)
+            self.checkpoint_files = sorted(
+                self.checkpoint_files, key=lambda x: x[1],
+                reverse=not self.decreasing)  # sort in descending order if a lower metric is not better
+
+            if self.verbose:
+                print("Current checkpoints:")
+                for c in self.checkpoint_files:
+                    print(c)
+
+            if metric is not None and (self.best_metric is None or self.cmp(metric, self.best_metric)):
+                self.best_epoch = epoch
+                self.best_metric = metric
                 shutil.copyfile(save_path, os.path.join(self.checkpoint_dir, 'model_best' + self.extension))
-        return None, None if self.best_metric is None else self.best_metric
+
+        return (None, None) if self.best_metric is None else (self.best_metric, self.best_epoch)
 
     def _cleanup_checkpoints(self, trim=0):
         trim = min(len(self.checkpoint_files), trim)
@@ -62,7 +76,8 @@ class CheckpointSaver:
         to_delete = self.checkpoint_files[delete_index:]
         for d in to_delete:
             try:
-                print('Cleaning checkpoint: ', d)
+                if self.verbose:
+                    print('Cleaning checkpoint: ', d)
                 os.remove(d[0])
             except Exception as e:
                 print('Exception (%s) while deleting checkpoint' % str(e))
@@ -74,7 +89,8 @@ class CheckpointSaver:
         torch.save(state, save_path)
         if os.path.exists(self.last_recovery_file):
             try:
-                print('Cleaning recovery', self.last_recovery_file)
+                if self.verbose:
+                    print('Cleaning recovery', self.last_recovery_file)
                 os.remove(self.last_recovery_file)
             except Exception as e:
                 print("Exception (%s) while removing %s" % (str(e), self.last_recovery_file))
@@ -143,8 +159,8 @@ def get_outdir(path, *paths, inc=False):
 
 def update_summary(epoch, train_metrics, eval_metrics, filename, write_header=False):
     rowd = OrderedDict(epoch=epoch)
-    rowd.update(train_metrics)
-    rowd.update(eval_metrics)
+    rowd.update([('train_' + k, v) for k, v in train_metrics.items()])
+    rowd.update([('eval_' + k, v) for k, v in eval_metrics.items()])
     with open(filename, mode='a') as cf:
         dw = csv.DictWriter(cf, fieldnames=rowd.keys())
         if write_header:  # first iteration (epoch == 1 can't be used)