tests on classification results

app/app.py

@@ -41,9 +41,10 @@ async def rerank_items(input_data: RankingRequest) -> RankingResponse:
     items = input_data.items
     # TODO consider sampling them?
 
-    print(items)
     reranked_ids, first_topic, insertion_pos = redistribute(platform=platform, items=items)
     #reranked_ids = [ for id_ in reranked_ids]
+    print("Receiving boost on: ", first_topic)
+    print("Position: ", insertion_pos)
 
     user_in_db = user_db.get_user(user_id=user)
 
@@ -97,6 +98,5 @@ async def rerank_items(input_data: RankingRequest) -> RankingResponse:
 
         # no civic content to boost on
         else:
-            print("there")
             return RankingResponse(ranked_ids=reranked_ids, new_items=[])        
 

app/modules/classify.py

@@ -10,7 +10,7 @@ except:
     print("No GPU available, running on CPU")
     device = None
 
-#model = pipeline("zero-shot-classification", model="facebook/bart-large-mnli")
+#model = pipeline("zero-shot-classification", model="facebook/bart-large-mnli", device=device)
 model = pipeline("zero-shot-classification", model="valhalla/distilbart-mnli-12-9", device=device)
 
 label_map = {
@@ -49,6 +49,7 @@ def classify(texts: List[str], labels: List[str]):
     # Iterate through each text to check for special cases
     for index, text in enumerate(texts):
         if text == "NON-VALID":
+            print("NON-VALID TEXT!!", text)
             # If text is "X", directly assign the label and score
             results.append({
                 "sequence": text,
@@ -57,16 +58,16 @@ def classify(texts: List[str], labels: List[str]):
             })
         else:
             # Otherwise, prepare for model processing
+            #print("- text =>", text)
             model_texts.append(text)
             model_indices.append(index)
     
     if model_texts:
         # Process texts through the model if there are any
-        predicted_labels = model(model_texts, labels, multi_label=False, batch_size=16)
+        predicted_labels = model(model_texts, labels, multi_label=False, batch_size=32)
         
         # Insert model results into the correct positions
         for pred, idx in zip(predicted_labels, model_indices):
             results.insert(idx, pred)
-    
-    print(results)
+    print([(r['labels'][0], r['sequence']) for r in results])
     return results

app/modules/redistribute.py

@@ -24,9 +24,7 @@ def redistribute(platform, items):
     mapped_scores = map_scores(predicted_labels=predicted_labels, default_label="something else")
     first_topic, insertion_pos = get_first_relevant_label(predicted_labels=predicted_labels, mapped_scores=mapped_scores, default_label="something else")
     # TODO include parent linking
-    print("OK--", predicted_labels)
     reranked_ids, _ = distribute_evenly(ids=[item.id for item in items], scores=mapped_scores)
-    print(reranked_ids)
     return reranked_ids, first_topic, insertion_pos