added option to choose postprocessing

app.py

@@ -29,7 +29,17 @@ def choose_model():
             return model_cp, model_cp
 
 
-def postprocess(resp):
+def choose_postprocessing():
+    with st.sidebar:
+        st.write('# Postprocessing')
+        postprocessing = st.selectbox('Select postprocessing method',
+            ('no postprocessing',
+             'remove substrings',
+             'iteration'))
+        return postprocessing
+
+
+def remove_substring(resp):
   """Function filters postprocessed model output to remove answers that
   are substrings of another answer."""
   answers = [answer['answer'] for answer in resp]
@@ -44,6 +54,32 @@ def postprocess(resp):
   return filtered_resp
 
 
+def iterate(resp, context):
+    # Remove substring answers
+    filtered_resp = remove_substring(resp)
+    # Remove best answer from original context
+    best_answer = filtered_resp[0]['answer']
+    new_context = context.replace(best_answer, "")
+    return new_context, filtered_resp
+
+
+def remove_substring_iter(pipeline, question, context):
+    # Create answers placeholder
+    ret_resp = []
+
+    # Loop through five times, removing best and re-running
+    for _ in range(5):
+        resp = pipeline(question, context, top_k=5,
+                        handle_impossible_answer=True)
+        # Update context
+        context, filtered_resp = iterate(resp, context)
+        # If best score not above threshold, quit
+        if filtered_resp[0]['score'] < 1e-3:
+            break
+        ret_resp.append(filtered_resp[0])
+    return ret_resp
+
+
 # Page config
 title = "Recipe Improver"
 icon = "🍣"
@@ -51,8 +87,12 @@ st.set_page_config(page_title=title, page_icon=icon)
 st.title(title)
 
 
-# Load model and tokenzier
+# Choose model and postprocessing procedure
 model_cp, tokenizer_cp = choose_model()
+postprocessing = choose_postprocessing()
+
+
+# Load model and tokenizer
 question_answer = load_pipeline(model_cp, tokenizer_cp)
 st.write("Model and tokenizer successfully loaded.")
 
@@ -70,10 +110,20 @@ if uploaded_file is not None:
     user_idx = st.text_input('Enter index')
     first_example = df['review'][int(user_idx)]
     question = "How was the recipe improved?"
-    resp = question_answer(question=question,
+
+    if postprocessing == "no postprocessing":
+        resp = question_answer(question=question,
+                             context=first_example,
+                             top_k=5)
+    elif postprocessing == "remove substrings":
+        resp = question_answer(question=question,
                              context=first_example,
                              top_k=5)
-    filtered_resp = postprocess(resp)
+        resp = remove_substring(resp)
+    elif postprocessing == "iteration":
+        resp = remove_substring_iter(question_answer,
+                                     question,
+                                     first_example)
 
     # Present results
     st.markdown(f"""
@@ -83,16 +133,11 @@ if uploaded_file is not None:
 
     {first_example}
 
-    The question was:
+    The question asked was:
 
     {question}
 
-    The unfiltered answers were:
+    The answers were:
 
     {resp}
-
-    The filtered answers were:
-
-    {filtered_resp}
-
     """)