Specify default train_len for iterator dataset

pytorch_pfn_extras/training/_trainer.py

@@ -247,9 +247,15 @@ def run(
             - :meth:`pytorch_pfn_extras.training._evaluator.Evaluator`
         """
         if train_len is None:
-            train_len = len(train_loader)  # type: ignore[arg-type]
+            if hasattr(train_loader, "__len__"):
+                train_len = len(train_loader)  # type: ignore[arg-type]
+            else:
+                train_len = 1
         if eval_len is None and val_loader is not None:
-            eval_len = len(val_loader)  # type: ignore[arg-type]
+            if hasattr(eval_len, "__len__"):
+                eval_len = len(val_loader)  # type: ignore[arg-type]
+            else:
+                eval_len = 1
 
         self._train_len = train_len
         self._eval_len = eval_len

tests/pytorch_pfn_extras_tests/training_tests/test_trainer.py

@@ -948,3 +948,39 @@ def test_create_distributed_evaluator():
     with mock.patch.object(dist, "is_initialized", return_value=True):
         evaluator = engine.create_evaluator(models=model, distributed=True)
     assert isinstance(evaluator, DistributedEvaluator)
+
+
+def test_trainer_run_with_iterator():
+    model = MyModel()
+    model = MyModelWithLossFn(model)
+    model = ppe.to(model, "cpu")
+    optimier = mock.MagicMock(spec=torch.optim.Optimizer)
+    evaluator = engine.create_evaluator(models=model)
+    trainer = engine.create_trainer(
+        models=model, optimizers=optimier, max_epochs=1, evaluator=evaluator
+    )
+    train_iterator = (
+        {
+            "x": torch.rand(
+                20,
+            ),
+            "t": torch.rand(
+                10,
+            ),
+        }
+        for i in range(10)
+    )
+    valid_iterator = (
+        {
+            "x": torch.rand(
+                20,
+            ),
+            "t": torch.rand(
+                10,
+            ),
+        }
+        for i in range(5)
+    )
+    trainer.run(train_iterator, valid_iterator)
+    assert trainer._train_len == 1
+    assert trainer._eval_len == 1