various improvements

README.md

@@ -5,7 +5,7 @@ colorFrom: purple
 colorTo: blue
 sdk: streamlit
 sdk_version: 1.10.0
-app_file: rock_fact_checker.py
+app_file: Rock_fact_checker.py
 pinned: false
 license: apache-2.0
 ---

Rock_fact_checker.py

@@ -0,0 +1,117 @@
+import streamlit as st
+
+import time
+import streamlit as st
+import logging
+from json import JSONDecodeError
+# from markdown import markdown
+# from annotated_text import annotation
+# from urllib.parse import unquote
+import random
+
+from app_utils.backend_utils import load_questions, query
+from app_utils.frontend_utils import set_state_if_absent, reset_results 
+from app_utils.config import RETRIEVER_TOP_K
+
+
+def main():
+
+
+    questions = load_questions()
+
+    # Persistent state
+    set_state_if_absent('question', "Elvis Presley is alive")
+    set_state_if_absent('answer', '')
+    set_state_if_absent('results', None)
+    set_state_if_absent('raw_json', None)
+    set_state_if_absent('random_question_requested', False)
+
+
+    ## MAIN CONTAINER
+    st.write("# Fact checking 🎸 Rocks!")
+    st.write()
+    st.markdown("""
+    ##### Enter a factual statement about [Rock music](https://en.wikipedia.org/wiki/List_of_mainstream_rock_performers) and let the AI check it out for you...
+    """)
+    # Search bar
+    question = st.text_input("", value=st.session_state.question,
+                             max_chars=100, on_change=reset_results)
+    col1, col2 = st.columns(2)
+    col1.markdown(
+        "<style>.stButton button {width:100%;}</style>", unsafe_allow_html=True)
+    col2.markdown(
+        "<style>.stButton button {width:100%;}</style>", unsafe_allow_html=True)
+    # Run button
+    run_pressed = col1.button("Run")
+    # Random question button
+    if col2.button("Random question"):
+        reset_results()
+        question = random.choice(questions)
+        # Avoid picking the same question twice (the change is not visible on the UI)
+        while question == st.session_state.question:
+            question = random.choice(questions)
+        st.session_state.question = question
+        st.session_state.random_question_requested = True
+        # Re-runs the script setting the random question as the textbox value
+        # Unfortunately necessary as the Random Question button is _below_ the textbox
+        # raise st.script_runner.RerunException(
+        #     st.script_request_queue.RerunData(None))
+    else:
+        st.session_state.random_question_requested = False
+    run_query = (run_pressed or question != st.session_state.question) \
+        and not st.session_state.random_question_requested
+
+    # Get results for query
+    if run_query and question:
+        time_start = time.time()
+        reset_results()
+        st.session_state.question = question
+        with st.spinner("🧠 &nbsp;&nbsp; Performing neural search on documents..."):
+            try:
+                st.session_state.results = query(
+                    question, RETRIEVER_TOP_K)
+                time_end = time.time()
+                print(time.strftime("%Y-%m-%d %H:%M:%S", time.gmtime()))
+                print(f'elapsed time: {time_end - time_start}')
+            except JSONDecodeError as je:
+                st.error(
+                    "👓 &nbsp;&nbsp; An error occurred reading the results. Is the document store working?")
+                return
+            except Exception as e:
+                logging.exception(e)
+                st.error("🐞 &nbsp;&nbsp; An error occurred during the request.")
+                return
+
+    # # Display results
+    # if st.session_state.results:
+    #     st.write("## Results:")
+    #     alert_irrelevance = True
+    #     if len(st.session_state.results['answers']) == 0:
+    #         st.info("""🤔 &nbsp;&nbsp; Haystack is unsure whether any of 
+    # the documents contain an answer to your question. Try to reformulate it!""")
+
+    #     for result in st.session_state.results['answers']:
+    #         result = result.to_dict()
+    #         if result["answer"]:
+    #             if alert_irrelevance and result['score'] < LOW_RELEVANCE_THRESHOLD:
+    #                 alert_irrelevance = False
+    #                 st.write("""
+    #                 <h4 style='color: darkred'>Attention, the 
+    #                 following answers have low relevance:</h4>""",
+    #                          unsafe_allow_html=True)
+
+    #         answer, context = result["answer"], result["context"]
+    #         start_idx = context.find(answer)
+    #         end_idx = start_idx + len(answer)
+    #         # Hack due to this bug: https://github.com/streamlit/streamlit/issues/3190
+    #         st.write(markdown("- ..."+context[:start_idx] +
+    #                 str(annotation(answer, "ANSWER", "#3e1c21", "white")) + 
+    #                 context[end_idx:]+"..."), unsafe_allow_html=True)
+    #         source = ""
+    #         name = unquote(result['meta']['name']).replace('_', ' ')
+    #         url = result['meta']['url']
+    #         source = f"[{name}]({url})"
+    #         st.markdown(
+    #             f"**Score:** {result['score']:.2f} -  **Source:** {source}")
+
+main()

app_utils/backend_utils.py

@@ -0,0 +1,58 @@
+import shutil
+from haystack.document_stores import FAISSDocumentStore
+from haystack.nodes import EmbeddingRetriever
+from haystack.pipelines import Pipeline
+
+import streamlit as st
+
+from app_utils.entailment_checker import EntailmentChecker
+
+from app_utils.config import STATEMENTS_PATH, INDEX_DIR, RETRIEVER_MODEL, RETRIEVER_MODEL_FORMAT, NLI_MODEL
+
+# cached to make index and models load only at start
+@st.cache(hash_funcs={"builtins.SwigPyObject": lambda _: None}, allow_output_mutation=True)
+def start_haystack():
+    """
+    load document store, retriever, reader and create pipeline
+    """
+    shutil.copy(f'{INDEX_DIR}/faiss_document_store.db', '.')
+    document_store = FAISSDocumentStore(
+        faiss_index_path=f'{INDEX_DIR}/my_faiss_index.faiss',
+        faiss_config_path=f'{INDEX_DIR}/my_faiss_index.json')
+    print(f'Index size: {document_store.get_document_count()}')
+    
+    retriever = EmbeddingRetriever(
+        document_store=document_store,
+        embedding_model=RETRIEVER_MODEL,
+        model_format=RETRIEVER_MODEL_FORMAT
+    )
+    
+    entailment_checker = EntailmentChecker(model_name_or_path=NLI_MODEL,
+                        use_gpu=False)
+    
+
+    pipe = Pipeline()
+    pipe.add_node(component=retriever, name="retriever", inputs=["Query"])
+    pipe.add_node(component=entailment_checker, name="ec", inputs=["retriever"])
+    return pipe
+
+pipe = start_haystack()
+# the pipeline is not included as parameter of the following function,
+# because it is difficult to cache
+@st.cache(persist=True, allow_output_mutation=True)
+def query(question: str, retriever_top_k: int = 5):
+    """Run query and get answers"""
+    params = {"retriever": {"top_k": retriever_top_k}}
+    results = pipe.run(question, params=params)
+    print(results)
+    return results      
+
+@st.cache()
+def load_questions():
+    """Load statements from file"""
+    with open(STATEMENTS_PATH) as fin:
+        questions = [line.strip() for line in fin.readlines()
+                     if not line.startswith('#')]
+    return questions
+
+              

app_utils/config.py

@@ -0,0 +1,8 @@
+
+INDEX_DIR = 'data/index'
+STATEMENTS_PATH = 'data/statements.txt'
+
+RETRIEVER_MODEL = "sentence-transformers/msmarco-distilbert-base-tas-b"
+RETRIEVER_MODEL_FORMAT = "sentence_transformers"
+RETRIEVER_TOP_K = 5
+NLI_MODEL = "valhalla/distilbart-mnli-12-1"

app_utils/frontend_utils.py

@@ -0,0 +1,15 @@
+import streamlit as st
+
+
+def set_state_if_absent(key, value):
+    if key not in st.session_state:
+        st.session_state[key] = value
+
+# Small callback to reset the interface in case the text of the question changes
+def reset_results(*args):
+    st.session_state.answer = None
+    st.session_state.results = None
+    st.session_state.raw_json = None
+
+
+              

data/statements.txt

@@ -0,0 +1,5 @@
+Kurt Cobain died in 1994
+Kurt Cobain died in 2008
+Green Day are a heavy metal band
+Green Day are a punk rock band
+The Beatles' first album was released in 1985

pages/Info.py

@@ -1,3 +1,3 @@
 import streamlit as st
 
-st.title("Test")
+