Update engine.py to resolve pytorch#1992

Update engine.py to resolve pytorch#1992

Author: puhuk

ignite/engine/engine.py

@@ -125,6 +125,10 @@ def compute_mean_std(engine, batch):
     _state_dict_all_req_keys = ("epoch_length", "max_epochs")
     _state_dict_one_of_opt_keys = ("iteration", "epoch")
 
+    DEBUG_EVENTS = 1
+    DEBUG_OUTPUT = 2
+    DEBUG_GRADS = 3
+
     # Flag to disable engine._internal_run as generator feature for BC
     interrupt_resume_enabled = True
 
@@ -425,6 +429,16 @@ def _fire_event(self, event_name: Any, *event_args: Any, **event_kwargs: Any) ->
             first, others = ((args[0],), args[1:]) if (args and args[0] == self) else ((), args)
             func(*first, *(event_args + others), **kwargs)
 
+        
+    def debug(self, level: int = 0, **kwargs):
+        if level > 2 :
+            self.logger.debug(f"{self.state.epoch} | {self.state.iteration}, Firing handlers for event {kwargs['event_name']}, Loss : {self.state.output}, LR : {kwargs['optimizer'].param_groups[0]['lr']}, Gradients : {kwargs['loss'].grad}")
+        elif level > 1 :
+            self.logger.debug(f"{self.state.epoch} | {self.state.iteration} Firing handlers for event {kwargs['event_name']}, Loss : {self.state.output}, LR : {kwargs['optimizer'].param_groups[0]['lr']}")
+        elif level > 0 :
+            self.logger.debug(f"{self.state.epoch} | {self.state.iteration}, Firing handlers for event {kwargs['event_name']}")
+
+
     def fire_event(self, event_name: Any) -> None:
         """Execute all the handlers associated with given event.