Epoch is implicitly incremented if terminated on iteration

🐛 Bug description

Below code shows the error:

from ignite.engine import Engine, Events
from ignite.utils import setup_logger

stop_iter = 50
epoch_length = 100
max_epochs = 2

trainer = Engine(lambda e, b: print(b, end=" "))
trainer.logger = setup_logger("trainer")
state = trainer.state

@trainer.on(Events.ITERATION_COMPLETED(every=stop_iter))
def stop():
    print("--> stop at {}".format(trainer.state.iteration))
    trainer.terminate()

data = list(range(epoch_length))

print("- Start from {} iteration".format(state.iteration))
state = trainer.run(data, max_epochs=max_epochs, epoch_length=epoch_length)
print("- Ended on {} iteration | {} epoch".format(state.iteration, state.epoch))

print("-- Do something else")

print("- Continue from {} iteration".format(state.iteration))
state = trainer.run(data, max_epochs=max_epochs, epoch_length=epoch_length)
print("- Ended on {} iteration | {} epoch".format(state.iteration, state.epoch))

print("-- Do something else")

print("- Continue from {} iteration".format(state.iteration))
state = trainer.run(data, max_epochs=max_epochs, epoch_length=epoch_length)
print("- Ended on {} iteration | {} epoch".format(state.iteration, state.epoch))

print("-- Do something else")

print("- Continue from {} iteration".format(state.iteration))
state = trainer.run(data, max_epochs=max_epochs, epoch_length=epoch_length)
print("- Ended on {} iteration | {} epoch".format(state.iteration, state.epoch))

print("-- Do something else")

print("- Continue from {} iteration".format(state.iteration))
state = trainer.run(data, max_epochs=max_epochs, epoch_length=epoch_length)
print("- Ended on {} iteration | {} epoch".format(state.iteration, state.epoch))

The issue is that iteration and epoch start to be unrelated which is a bug.

Environment

• PyTorch Version (e.g., 1.4):
• Ignite Version (e.g., 0.3.0):
• OS (e.g., Linux):
• How you installed Ignite (conda, pip, source):
• Python version:
• Any other relevant information:

Start by running the reproduction in the issue, then inspect the Engine.run flow and the trainer.terminate handling. Trace how iteration and epoch are updated when termination occurs at iteration 50. Done means repeated runs preserve a consistent relationship between the iteration and epoch values shown in the reproduction.