Update app.py

app.py

@@ -97,7 +97,7 @@ json_instructions='''
 def get_meta_score(url,question_embedding):
     qry = index.fetch(ids=[url], namespace="meta")
     try:
-        emb=qry['vectors'][resp['id']]['values']
+        emb=qry['vectors'][url]['values']
         vector1 = np.array(emb).reshape(1, -1)  # Reshape to ensure compatibility with sklearn
         vector2 = question_embedding.numpy().reshape(1, -1)
         
@@ -179,77 +179,64 @@ if QUESTION:
 
     ns='webpages'
     ns='full'
-    resp= query_from_pinecone(index,ns, question_embedding.tolist(), 3)
-    if len(resp)>0:
-        #st.write("Got results...extracting an answer")
-        out= resp[0]['metadata']['data']
-        try:
-            url= resp[0]['metadata']['url']
-        except:
-            url= resp[0]['id']
-        url= resp[0]['id']
-        
-        title=resp[0]['metadata']['title']
-        #+ '\n*************\n'+  resp[1]['metadata']['text'] + '\n*************\n'+ resp[2]['metadata']['text']
-        matches=[]
-        
+    resp= query_from_pinecone(index,ns, question_embedding.tolist(), 10)
+    resplist=[]
+    id=0
+    for r in resp:
+        id+=1
         d={}
-        d['id']=1
-        d['content']=resp[0]['metadata']['data']
-        d['url']=url
-        meta_score= get_meta_score(url,question_embedding)
-        score=.5* resp[0]['score'] + .5*meta_score
-        matches.append(d)
-
-        if len(resp)>1:
-            d['id']=2
-            d['content']=resp[1]['metadata']['data']
-            d['url']=url
-            meta_score= get_meta_score(url,question_embedding)
-            score=.5* resp[0]['score'] + .5*meta_score
-            matches.append(d)
-
-        sorted_indices = sorted(range(len(objects)), key=lambda i: objects[i]['score'], reverse=True)
-
-        # Get the elements with the top 2 highest values
-        top_2 =  [objects[i] for i in sorted_indices[:2]]
+        d['id']=id
+        d['content']=r['metadata']['data']
+        d['url']=r['id']
+        meta_score= get_meta_score(r['id'],question_embedding)
+        score=.5* r['score'] + .5*meta_score
+        d['score']=score
+        print(d['url'], score, r['score'], meta_score)
+        resplist.append(d)
+
+    sorted_indices = sorted(range(len(resplist)), key=lambda i: resplist[i]['score'], reverse=True)
+
+    # Get the elements with the top 2 highest values
+    top_2 =  [resplist[i] for i in sorted_indices[:2]]
             
-        # covert to array
-        
-        json_data = json.dumps(top_2)
+    # covert to array
+    
+    json_data = json.dumps(top_2)
 
-        st.write(json_data)
-        
-        if score>.5:
-
-            mode = "two"  # two passages
-
-            client = OpenAI()
-
-            if mode=="one":
-                instr=system_instructions_text
-                content="""
-                <text>
-                {}
-                </text>
-                """.format(out)
-                
-            if mode=="two":
-                instr=json_instructions
-                content=json_data
-
-            response = client.chat.completions.create(
-              model="gpt-3.5-turbo",
-              messages=[
-                {"role": "system", "content":instr },
-                {"role": "user", "content": content},
-                {"role": "user", "content": "Question:"+QUESTION}
-              ]
-            )
-
-            ans= response.choices[0].message.content
-        else:
-            ans='Weak match to your query. Please try reframing your question'
+    st.write(json_data)
+    goodmatch=False    
+    if resplist[sorted_indices[0]]['score']>.5:
+        goodmatch=True
+        mode = "two"  # two passages
+
+        client = OpenAI()
+
+        if mode=="one":
+            instr=system_instructions_text
+
+            out= resplist[sorted_indices[0]]['content']
+            content="""
+            <text>
+            {}
+            </text>
+            """.format(out)
+            
+        if mode=="two":
+            instr=json_instructions
+            content=json_data
+
+        response = client.chat.completions.create(
+          model="gpt-3.5-turbo",
+          messages=[
+            {"role": "system", "content":instr },
+            {"role": "user", "content": content},
+            {"role": "user", "content": "Question:"+QUESTION}
+          ]
+        )
+
+        ans= response.choices[0].message.content
+    else:
+        ans='Weak match to your query. Please try reframing your question'
 
         #st.write("Matched URL:{}  Score:{}".format(url,score))
         testing = False
@@ -261,6 +248,9 @@ if QUESTION:
 
         with st.chat_message("assistant"):
             response = st.write_stream(response_generator(ans))
+            if goodmatch:
+                st.write('Resources:'+top_2[0]['url'])
+                st.write(top_2[1]['url'])
             # Add assistant response to chat history
             st.session_state.messages.append({"role": "assistant", "content": response})
         #st.write(ans)