1052 Upgrade Ignite dependency to 0.4.2

docs/requirements.txt

@@ -1,6 +1,6 @@
 -f https://download.pytorch.org/whl/cpu/torch-1.4.0%2Bcpu-cp37-cp37m-linux_x86_64.whl
 torch>=1.4.0
-pytorch-ignite==0.3.0
+pytorch-ignite==0.4.2
 numpy>=1.17
 itk
 nibabel

monai/engines/evaluator.py

@@ -25,8 +25,8 @@
     from ignite.engine import Engine
     from ignite.metrics import Metric
 else:
-    Engine, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Engine")
-    Metric, _ = optional_import("ignite.metrics", "0.3.0", exact_version, "Metric")
+    Engine, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Engine")
+    Metric, _ = optional_import("ignite.metrics", "0.4.2", exact_version, "Metric")
 
 
 class Evaluator(Workflow):

monai/engines/multi_gpu_supervised_trainer.py

@@ -19,15 +19,15 @@
 from monai.engines.utils import get_devices_spec
 from monai.utils import exact_version, optional_import
 
-create_supervised_trainer, _ = optional_import("ignite.engine", "0.3.0", exact_version, "create_supervised_trainer")
-create_supervised_evaluator, _ = optional_import("ignite.engine", "0.3.0", exact_version, "create_supervised_evaluator")
-_prepare_batch, _ = optional_import("ignite.engine", "0.3.0", exact_version, "_prepare_batch")
+create_supervised_trainer, _ = optional_import("ignite.engine", "0.4.2", exact_version, "create_supervised_trainer")
+create_supervised_evaluator, _ = optional_import("ignite.engine", "0.4.2", exact_version, "create_supervised_evaluator")
+_prepare_batch, _ = optional_import("ignite.engine", "0.4.2", exact_version, "_prepare_batch")
 if TYPE_CHECKING:
     from ignite.engine import Engine
     from ignite.metrics import Metric
 else:
-    Engine, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Engine")
-    Metric, _ = optional_import("ignite.metrics", "0.3.0", exact_version, "Metric")
+    Engine, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Engine")
+    Metric, _ = optional_import("ignite.metrics", "0.4.2", exact_version, "Metric")
 
 
 def _default_transform(_x: torch.Tensor, _y: torch.Tensor, _y_pred: torch.Tensor, loss: torch.Tensor) -> float:

monai/engines/trainer.py

@@ -26,8 +26,8 @@
     from ignite.engine import Engine
     from ignite.metrics import Metric
 else:
-    Engine, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Engine")
-    Metric, _ = optional_import("ignite.metrics", "0.3.0", exact_version, "Metric")
+    Engine, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Engine")
+    Metric, _ = optional_import("ignite.metrics", "0.4.2", exact_version, "Metric")
 
 
 class Trainer(Workflow):
@@ -42,8 +42,6 @@ def run(self) -> None:
         If call this function multiple times, it will continuously run from the previous state.
 
         """
-        if self._is_done(self.state):
-            self.state.iteration = 0  # to avoid creating new State instance in ignite Engine.run
         self.scaler = torch.cuda.amp.GradScaler() if self.amp else None
         super().run()
 

monai/engines/workflow.py

@@ -19,15 +19,15 @@
 from monai.transforms import apply_transform
 from monai.utils import ensure_tuple, exact_version, optional_import
 
-IgniteEngine, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Engine")
-State, _ = optional_import("ignite.engine", "0.3.0", exact_version, "State")
-Events, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Events")
+IgniteEngine, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Engine")
+State, _ = optional_import("ignite.engine", "0.4.2", exact_version, "State")
+Events, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Events")
 if TYPE_CHECKING:
     from ignite.engine import Engine
     from ignite.metrics import Metric
 else:
-    Engine, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Engine")
-    Metric, _ = optional_import("ignite.metrics", "0.3.0", exact_version, "Metric")
+    Engine, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Engine")
+    Metric, _ = optional_import("ignite.metrics", "0.4.2", exact_version, "Metric")
 
 
 class Workflow(IgniteEngine):  # type: ignore[valid-type, misc] # due to optional_import
@@ -99,7 +99,7 @@ def set_sampler_epoch(engine: Engine):
             iteration=0,
             epoch=0,
             max_epochs=max_epochs,
-            epoch_length=-1,
+            epoch_length=len(data_loader),
             output=None,
             batch=None,
             metrics={},
@@ -154,7 +154,7 @@ def run(self) -> None:
         Execute training, validation or evaluation based on Ignite Engine.
 
         """
-        super().run(data=self.data_loader, epoch_length=len(self.data_loader))
+        super().run(data=self.data_loader, max_epochs=self.state.max_epochs)
 
     def _iteration(self, engine: Engine, batchdata: Dict[str, torch.Tensor]):
         """

monai/handlers/checkpoint_loader.py

@@ -16,12 +16,12 @@
 
 from monai.utils import exact_version, optional_import
 
-Events, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Events")
-Checkpoint, _ = optional_import("ignite.handlers", "0.3.0", exact_version, "Checkpoint")
+Events, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Events")
+Checkpoint, _ = optional_import("ignite.handlers", "0.4.2", exact_version, "Checkpoint")
 if TYPE_CHECKING:
     from ignite.engine import Engine
 else:
-    Engine, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Engine")
+    Engine, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Engine")
 
 
 class CheckpointLoader:

monai/handlers/checkpoint_saver.py

@@ -14,12 +14,12 @@
 
 from monai.utils import exact_version, optional_import
 
-Events, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Events")
-ModelCheckpoint, _ = optional_import("ignite.handlers", "0.3.0", exact_version, "ModelCheckpoint")
+Events, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Events")
+ModelCheckpoint, _ = optional_import("ignite.handlers", "0.4.2", exact_version, "ModelCheckpoint")
 if TYPE_CHECKING:
     from ignite.engine import Engine
 else:
-    Engine, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Engine")
+    Engine, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Engine")
 
 
 class CheckpointSaver:

monai/handlers/classification_saver.py

@@ -15,11 +15,11 @@
 from monai.data import CSVSaver
 from monai.utils import exact_version, optional_import
 
-Events, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Events")
+Events, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Events")
 if TYPE_CHECKING:
     from ignite.engine import Engine
 else:
-    Engine, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Engine")
+    Engine, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Engine")
 
 
 class ClassificationSaver:

monai/handlers/lr_schedule_handler.py

@@ -16,11 +16,11 @@
 
 from monai.utils import ensure_tuple, exact_version, optional_import
 
-Events, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Events")
+Events, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Events")
 if TYPE_CHECKING:
     from ignite.engine import Engine
 else:
-    Engine, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Engine")
+    Engine, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Engine")
 
 
 class LrScheduleHandler:

monai/handlers/mean_dice.py

@@ -16,10 +16,10 @@
 from monai.metrics import DiceMetric
 from monai.utils import MetricReduction, exact_version, optional_import
 
-NotComputableError, _ = optional_import("ignite.exceptions", "0.3.0", exact_version, "NotComputableError")
-Metric, _ = optional_import("ignite.metrics", "0.3.0", exact_version, "Metric")
-reinit__is_reduced, _ = optional_import("ignite.metrics.metric", "0.3.0", exact_version, "reinit__is_reduced")
-sync_all_reduce, _ = optional_import("ignite.metrics.metric", "0.3.0", exact_version, "sync_all_reduce")
+NotComputableError, _ = optional_import("ignite.exceptions", "0.4.2", exact_version, "NotComputableError")
+Metric, _ = optional_import("ignite.metrics", "0.4.2", exact_version, "Metric")
+reinit__is_reduced, _ = optional_import("ignite.metrics.metric", "0.4.2", exact_version, "reinit__is_reduced")
+sync_all_reduce, _ = optional_import("ignite.metrics.metric", "0.4.2", exact_version, "sync_all_reduce")
 
 
 class MeanDice(Metric):  # type: ignore[valid-type, misc] # due to optional_import

monai/handlers/metric_logger.py

@@ -14,11 +14,11 @@
 
 from monai.utils import exact_version, optional_import
 
-Events, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Events")
+Events, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Events")
 if TYPE_CHECKING:
     from ignite.engine import Engine
 else:
-    Engine, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Engine")
+    Engine, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Engine")
 
 
 class MetricLogger:

monai/handlers/roc_auc.py

@@ -17,8 +17,8 @@
 from monai.metrics import compute_roc_auc
 from monai.utils import Average, exact_version, optional_import
 
-Metric, _ = optional_import("ignite.metrics", "0.3.0", exact_version, "Metric")
-reinit__is_reduced, _ = optional_import("ignite.metrics.metric", "0.3.0", exact_version, "reinit__is_reduced")
+Metric, _ = optional_import("ignite.metrics", "0.4.2", exact_version, "Metric")
+reinit__is_reduced, _ = optional_import("ignite.metrics.metric", "0.4.2", exact_version, "reinit__is_reduced")
 
 
 class ROCAUC(Metric):  # type: ignore[valid-type, misc]  # due to optional_import

monai/handlers/segmentation_saver.py

@@ -17,11 +17,11 @@
 from monai.data import NiftiSaver, PNGSaver
 from monai.utils import GridSampleMode, GridSamplePadMode, InterpolateMode, exact_version, optional_import
 
-Events, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Events")
+Events, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Events")
 if TYPE_CHECKING:
     from ignite.engine import Engine
 else:
-    Engine, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Engine")
+    Engine, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Engine")
 
 
 class SegmentationSaver:

monai/handlers/smartcache_handler.py

@@ -14,11 +14,11 @@
 from monai.data import SmartCacheDataset
 from monai.utils import exact_version, optional_import
 
-Events, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Events")
+Events, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Events")
 if TYPE_CHECKING:
     from ignite.engine import Engine
 else:
-    Engine, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Engine")
+    Engine, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Engine")
 
 
 class SmartCacheHandler:

monai/handlers/stats_handler.py

@@ -17,11 +17,11 @@
 
 from monai.utils import exact_version, is_scalar, optional_import
 
-Events, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Events")
+Events, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Events")
 if TYPE_CHECKING:
     from ignite.engine import Engine
 else:
-    Engine, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Engine")
+    Engine, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Engine")
 
 DEFAULT_KEY_VAL_FORMAT = "{}: {:.4f} "
 DEFAULT_TAG = "Loss"

monai/handlers/tensorboard_handlers.py

@@ -18,12 +18,12 @@
 from monai.utils import exact_version, is_scalar, optional_import
 from monai.visualize import plot_2d_or_3d_image
 
-Events, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Events")
+Events, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Events")
 if TYPE_CHECKING:
     from ignite.engine import Engine
     from torch.utils.tensorboard import SummaryWriter
 else:
-    Engine, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Engine")
+    Engine, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Engine")
     SummaryWriter, _ = optional_import("torch.utils.tensorboard", name="SummaryWriter")
 
 DEFAULT_TAG = "Loss"

monai/handlers/utils.py

@@ -16,7 +16,7 @@
 if TYPE_CHECKING:
     from ignite.engine import Engine
 else:
-    Engine, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Engine")
+    Engine, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Engine")
 
 
 def stopping_fn_from_metric(metric_name: str) -> Callable[[Engine], Any]:

monai/handlers/validation_handler.py

@@ -14,11 +14,11 @@
 from monai.engines.evaluator import Evaluator
 from monai.utils import exact_version, optional_import
 
-Events, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Events")
+Events, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Events")
 if TYPE_CHECKING:
     from ignite.engine import Engine
 else:
-    Engine, _ = optional_import("ignite.engine", "0.3.0", exact_version, "Engine")
+    Engine, _ = optional_import("ignite.engine", "0.4.2", exact_version, "Engine")
 
 
 class ValidationHandler:

requirements-dev.txt

@@ -1,6 +1,6 @@
 # Full requirements for developments
 -r requirements-min.txt
-pytorch-ignite==0.3.0
+pytorch-ignite==0.4.2
 gdown>=3.6.4
 scipy
 itk

setup.cfg

@@ -24,7 +24,7 @@ all =
     scikit-image>=0.14.2
     pillow
     tensorboard
-    pytorch-ignite==0.3.0
+    pytorch-ignite==0.4.2
     gdown>=3.6.4
     torchvision
     itk
@@ -39,7 +39,7 @@ tensorboard =
 gdown =
     gdown>=3.6.4
 ignite =
-    pytorch-ignite==0.3.0
+    pytorch-ignite==0.4.2
 torchvision =
     torchvision
 itk =

tests/test_handler_checkpoint_loader.py

@@ -32,11 +32,12 @@ def test_one_save_one_load(self):
         data2 = net2.state_dict()
         data2["weight"] = torch.tensor([0.2])
         net2.load_state_dict(data2)
-        engine = Engine(lambda e, b: None)
         with tempfile.TemporaryDirectory() as tempdir:
+            engine = Engine(lambda e, b: None)
             CheckpointSaver(save_dir=tempdir, save_dict={"net": net1}, save_final=True).attach(engine)
             engine.run([0] * 8, max_epochs=5)
-            path = tempdir + "/net_final_iteration=40.pth"
+            path = tempdir + "/net_final_iteration=40.pt"
+            engine = Engine(lambda e, b: None)
             CheckpointLoader(load_path=path, load_dict={"net": net2}).attach(engine)
             engine.run([0] * 8, max_epochs=1)
             torch.testing.assert_allclose(net2.state_dict()["weight"], 0.1)
@@ -52,12 +53,13 @@ def test_two_save_one_load(self):
         data2 = net2.state_dict()
         data2["weight"] = torch.tensor([0.2])
         net2.load_state_dict(data2)
-        engine = Engine(lambda e, b: None)
         with tempfile.TemporaryDirectory() as tempdir:
+            engine = Engine(lambda e, b: None)
             save_dict = {"net": net1, "opt": optimizer}
             CheckpointSaver(save_dir=tempdir, save_dict=save_dict, save_final=True).attach(engine)
             engine.run([0] * 8, max_epochs=5)
-            path = tempdir + "/checkpoint_final_iteration=40.pth"
+            path = tempdir + "/checkpoint_final_iteration=40.pt"
+            engine = Engine(lambda e, b: None)
             CheckpointLoader(load_path=path, load_dict={"net": net2}).attach(engine)
             engine.run([0] * 8, max_epochs=1)
             torch.testing.assert_allclose(net2.state_dict()["weight"], 0.1)
@@ -73,11 +75,12 @@ def test_save_single_device_load_multi_devices(self):
         data2["weight"] = torch.tensor([0.2])
         net2.load_state_dict(data2)
         net2 = torch.nn.DataParallel(net2)
-        engine = Engine(lambda e, b: None)
         with tempfile.TemporaryDirectory() as tempdir:
+            engine = Engine(lambda e, b: None)
             CheckpointSaver(save_dir=tempdir, save_dict={"net": net1}, save_final=True).attach(engine)
             engine.run([0] * 8, max_epochs=5)
-            path = tempdir + "/net_final_iteration=40.pth"
+            path = tempdir + "/net_final_iteration=40.pt"
+            engine = Engine(lambda e, b: None)
             CheckpointLoader(load_path=path, load_dict={"net": net2}).attach(engine)
             engine.run([0] * 8, max_epochs=1)
             torch.testing.assert_allclose(net2.state_dict()["module.weight"], 0.1)

tests/test_handler_checkpoint_saver.py

@@ -22,7 +22,7 @@
 
 from monai.handlers import CheckpointSaver
 
-TEST_CASE_1 = [True, False, None, 1, True, 0, None, ["test_checkpoint_final_iteration=40.pth"]]
+TEST_CASE_1 = [True, False, None, 1, True, 0, None, ["test_checkpoint_final_iteration=40.pt"]]
 
 TEST_CASE_2 = [
     False,
@@ -32,10 +32,10 @@
     True,
     0,
     None,
-    ["test_checkpoint_key_metric=32.pth", "test_checkpoint_key_metric=40.pth"],
+    ["test_checkpoint_key_metric=32.pt", "test_checkpoint_key_metric=40.pt"],
 ]
 
-TEST_CASE_3 = [False, False, None, 1, True, 2, 2, ["test_checkpoint_epoch=2.pth", "test_checkpoint_epoch=4.pth"]]
+TEST_CASE_3 = [False, False, None, 1, True, 2, 2, ["test_checkpoint_epoch=2.pt", "test_checkpoint_epoch=4.pt"]]
 
 TEST_CASE_4 = [
     False,
@@ -45,10 +45,10 @@
     False,
     10,
     2,
-    ["test_checkpoint_iteration=30.pth", "test_checkpoint_iteration=40.pth"],
+    ["test_checkpoint_iteration=30.pt", "test_checkpoint_iteration=40.pt"],
 ]
 
-TEST_CASE_5 = [True, False, None, 1, True, 0, None, ["test_checkpoint_final_iteration=40.pth"], True]
+TEST_CASE_5 = [True, False, None, 1, True, 0, None, ["test_checkpoint_final_iteration=40.pt"], True]
 
 
 class TestHandlerCheckpointSaver(unittest.TestCase):
@@ -115,7 +115,7 @@ def _train_func(engine, batch):
 
             with self.assertRaises(RuntimeError):
                 engine.run(range(3), max_epochs=2)
-            self.assertTrue(os.path.exists(os.path.join(tempdir, "net_final_iteration=1.pth")))
+            self.assertTrue(os.path.exists(os.path.join(tempdir, "net_final_iteration=1.pt")))
 
 
 if __name__ == "__main__":