Make seed setting more flexible - BC breaking change

ignite/engine/engine.py

@@ -564,8 +564,8 @@ def run(
             epoch_length (int, optional): Number of iterations to count as one epoch. By default, it can be set as
                 `len(data)`. If `data` is an iterator and `epoch_length` is not set, an error is raised.
                 This argument should be `None` if run is resuming from a state.
-            seed (int, optional): Seed to setup at each epoch for reproducible runs.
-                This argument should be `None` if run is resuming from a state.
+            seed (int, optional): Seed to use for dataflow consistency, by default it
+                will respect the global random state. This argument should be `None` if run is resuming from a state.
 
         Returns:
             State: output state.
@@ -594,7 +594,7 @@ def switch_batch(engine):
             if max_epochs is None:
                 max_epochs = 1
             if seed is None:
-                seed = 12
+                seed = torch.randint(0, int(1e9), (1,)).item()
             if epoch_length is None:
                 if hasattr(data, "__len__"):
                     epoch_length = len(data)

tests/ignite/engine/test_engine.py

@@ -6,6 +6,7 @@
 import torch
 
 from ignite.engine import Engine, Events, State
+from ignite.metrics import Average
 
 
 def test_terminate():
@@ -529,3 +530,35 @@ def foo(e, b):
     engine.run(data_loader, epoch_length=10, max_epochs=5)
 
     assert counter[0] == 50
+
+
+def test_engine_random_state():
+    def random_data_generator():
+        while True:
+            yield torch.randint(0, 100, size=(5,))
+
+    def sum_data(engine, batch):
+        result = torch.sum(batch)
+        return result
+
+    def get_engine():
+        engine = Engine(sum_data)
+        average = Average()
+        average.attach(engine, "average")
+        return engine
+
+    torch.manual_seed(34)
+    engine = get_engine()
+    state1 = engine.run(random_data_generator(), max_epochs=2, epoch_length=2)
+
+    torch.manual_seed(34)
+    engine = get_engine()
+    state2 = engine.run(random_data_generator(), max_epochs=2, epoch_length=2)
+
+    torch.manual_seed(42)
+    engine = get_engine()
+    state3 = engine.run(random_data_generator(), max_epochs=2, epoch_length=2)
+
+    assert state1.metrics["average"] == pytest.approx(state2.metrics["average"])
+    assert state1.metrics["average"] != pytest.approx(state3.metrics["average"])
+    assert state2.metrics["average"] != pytest.approx(state3.metrics["average"])