some fixes to WalkRouting

src/decentralizepy/communication/TCPRandomWalkRouting.py

@@ -561,7 +561,7 @@ class TCPRandomWalkRoutingInternal(TCPRandomWalkBase):
         if self.current_data == None:  # have not yet received this rounds data
             return
         if len(self.outgoing_request) != 0:
-            logging.debug(f"still have outgoing requests {len(self.outgoing_request)}")
+            logging.debug(f"still have outgoing requests {self.outgoing_request}")
             return
         if len(self.future_neighbours.get(self.current_round, [])) != 0:
             logging.debug(
@@ -649,7 +649,7 @@ class TCPRandomWalkRoutingInternal(TCPRandomWalkBase):
                     self.send(
                         sender, self.current_data
                     )  # messaging is tcp based so this should always arrive after helo
-            elif round == self.current_round + 1:  # they are ahead
+            elif round > self.current_round:  # they are ahead
                 self.future_neighbours.setdefault(round, []).append((sender, False))
                 logging.info(f"Added {sender} to future neighbors")
                 # we conntact them in the next local round
@@ -812,6 +812,7 @@ class TCPRandomWalkRoutingInternal(TCPRandomWalkBase):
                         f"Response to our disconnect request for {sender} with round {bye[1]}"
                     )
             else:  # this goodbye was not initiated by us
+                logging.debug(f"sender {sender} not in outgoing byes")
                 all_to_be_removed = self.this_round_bye.union(
                     set(self.future_byes.keys())
                 )

src/decentralizepy/datasets/CIFAR10.py

@@ -210,7 +210,7 @@ class CIFAR10(Dataset):
                 count += 1
                 _, predictions = torch.max(outputs, 1)
                 for label, prediction in zip(labels, predictions):
-                    logging.debug("{} predicted as {}".format(label, prediction))
+                    # logging.debug("{} predicted as {}".format(label, prediction))
                     if label == prediction:
                         correct_pred[label] += 1
                         total_correct += 1
@@ -224,7 +224,7 @@ class CIFAR10(Dataset):
                 accuracy = 100 * float(value) / total_pred[key]
             else:
                 accuracy = 100.0
-            logging.debug("Accuracy for class {} is: {:.1f} %".format(key, accuracy))
+            # logging.debug("Accuracy for class {} is: {:.1f} %".format(key, accuracy))
 
         accuracy = 100 * float(total_correct) / total_predicted
         loss_val = loss_val / count