Update app.py

app.py

@@ -20,36 +20,52 @@ st.title("Economics & Population Advisor")
 uploaded_file = st.file_uploader("Upload your CSV file with economic documents", type=["csv"])
 
 if uploaded_file is not None:
-    # Load CSV
-    df = pd.read_csv(uploaded_file, error_bad_lines=False, engine='python')
+    # Load CSV with error handling
+    df = pd.read_csv(uploaded_file, on_bad_lines='skip', engine='python')
     st.write("Dataset Preview:", df.head())
 
-    # Assume 'text' column contains the document text; replace with actual column name
-    documents = df['Country Name'].tolist() if 'text' in df.columns else st.text_input("Specify the text column name:")
-
-    # Create embeddings for FAISS indexing
-    st.write("Indexing documents...")
-    embeddings = embedding_model.encode(documents)
-    dimension = embeddings.shape[1]
-    index = faiss.IndexFlatL2(dimension)
-    index.add(np.array(embeddings))
-    st.write("Indexing complete.")
-
-    # Function to generate response
-    def generate_summary(context):
-        inputs = tokenizer("summarize: " + context, return_tensors="pt", max_length=512, truncation=True)
-        outputs = qa_model.generate(inputs["input_ids"], max_length=150, min_length=50, length_penalty=2.0)
-        return tokenizer.decode(outputs[0], skip_special_tokens=True)
-
-    # RAG functionality: Ask a question, retrieve documents, and generate an answer
-    st.subheader("Ask a Question about Economic Data")
-    question = st.text_input("Enter your question:")
-
-    if st.button("Get Answer") and question:
-        question_embedding = embedding_model.encode([question])
-        D, I = index.search(np.array(question_embedding), k=3)
-        retrieved_docs = [documents[i] for i in I[0]]
-        context = " ".join(retrieved_docs)
-        answer = generate_summary(context)
+    # Allow user to specify the column containing the text (economic documents)
+    text_column = st.text_input("Specify the column containing the document text:", value="Country Name")
+
+    if text_column not in df.columns:
+        st.error(f"The column '{text_column}' was not found in the dataset.")
+    else:
+        # Extract documents from the specified column
+        documents = df[text_column].tolist()
+
+        # Create embeddings for FAISS indexing
+        st.write("Indexing documents...")
+        embeddings = embedding_model.encode(documents, convert_to_numpy=True)
+        dimension = embeddings.shape[1]
         
-        st.write("Answer:", answer)
+        # Create a FAISS index and add embeddings
+        index = faiss.IndexFlatL2(dimension)
+        index.add(np.array(embeddings, dtype=np.float32))
+        st.write("Indexing complete.")
+
+        # Function to generate summary using T5 model
+        def generate_summary(context):
+            inputs = tokenizer("summarize: " + context, return_tensors="pt", max_length=512, truncation=True)
+            outputs = qa_model.generate(inputs["input_ids"], max_length=150, min_length=50, length_penalty=2.0)
+            return tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+        # RAG functionality: Ask a question, retrieve documents, and generate an answer
+        st.subheader("Ask a Question about Economic Data")
+        question = st.text_input("Enter your question:")
+
+        if st.button("Get Answer") and question:
+            # Embed the question
+            question_embedding = embedding_model.encode([question], convert_to_numpy=True)
+
+            # Search for the most relevant documents
+            D, I = index.search(np.array(question_embedding, dtype=np.float32), k=3)
+            retrieved_docs = [documents[i] for i in I[0]]
+
+            # Combine retrieved documents into context
+            context = " ".join(retrieved_docs[:5])  # Limit to 5 documents to avoid long input
+            if len(context) > 1000:  # Truncate context if too long
+                context = context[:1000]
+
+            # Generate summary using the context
+            answer = generate_summary(context)
+            st.write("Answer:", answer)