Aktualisieren models.py

models.py

@@ -4,6 +4,8 @@ from torch import nn
 import numpy as np
 
 import utils #
+import sys
+import traceback
 
 
 class SummarisationModel(nn.Module):
@@ -120,23 +122,29 @@ class ActorOnlySummarisationModel(SummarisationModel):
             self.optimizer.zero_grad()
 
             for datapoint in batch:
-                # documents with empty content!
-                if len(datapoint.raw_document) == 0 or len(datapoint.raw_summary) == 0: 
-                    print("Warning! This datapoint has an empty document or an empty summary")
-                    continue
 
-                _, probs = self.__call__(datapoint.document)
+                try:    # Prevent breakdown for inapt datapoints
+                    # documents with empty content!
+                    if len(datapoint.raw_document) == 0 or len(datapoint.raw_summary) == 0: 
+                        print("Warning! This datapoint has an empty document or an empty summary")
+                        continue
 
-                o = datapoint.p_searchspace @ torch.log(probs) + datapoint.n_searchspace @ torch.log(1 - probs)
+                    _, probs = self.__call__(datapoint.document)
 
-                idx_sample = torch.argmax(o)
+                    o = datapoint.p_searchspace @ torch.log(probs) + datapoint.n_searchspace @ torch.log(1 - probs)
 
-                loss = - datapoint.top_rouge[idx_sample] * o[idx_sample]
+                    idx_sample = torch.argmax(o)
 
-                loss.backward()
+                    loss = - datapoint.top_rouge[idx_sample] * o[idx_sample]
 
-                epoch_loss += loss.item()
-                epoch_rouge += datapoint.top_rouge[idx_sample]
+                    loss.backward()
+
+                    epoch_loss += loss.item()
+                    epoch_rouge += datapoint.top_rouge[idx_sample]
+                
+                except Exception as e:
+                    traceback.print_exception(*sys.exc_info())
+                    continue
             
             self.optimizer.step()