app.py

import gradio as gr
import pandas as pd

from haystack.schema import Answer
from haystack.document_stores import InMemoryDocumentStore
from haystack.pipelines import FAQPipeline, ExtractiveQAPipeline
from haystack.nodes import EmbeddingRetriever, TfidfRetriever, FARMReader, TextConverter, PreProcessor
from haystack.utils import print_answers
from haystack.utils import convert_files_to_docs
import logging

# FAQ Haystack function calls 

def start_haystack():
    document_store = InMemoryDocumentStore(index="document", embedding_field='embedding', embedding_dim=384, similarity='cosine')
    retriever = EmbeddingRetriever(document_store=document_store, embedding_model='sentence-transformers/all-MiniLM-L6-v2', use_gpu=True, top_k=1)
    load_data_to_store(document_store,retriever)
    pipeline = FAQPipeline(retriever=retriever)
    return pipeline

def load_data_to_store(document_store, retriever):
    df = pd.read_csv('monopoly_qa-v1.csv')
    questions = list(df.Question)
    df['embedding'] = retriever.embed_queries(texts=questions)
    df = df.rename(columns={"Question":"content","Answer":"answer"})
    df.drop('link to source (to prevent duplicate sources)',axis=1, inplace=True)
    
    dicts = df.to_dict(orient="records")
    document_store.write_documents(dicts)
    
faq_pipeline = start_haystack()
 
def predict_faq(question):
    prediction = faq_pipeline.run(question)
    answer =  prediction["answers"][0].meta
    faq_response = "FAQ Question: " + answer["query"] + "\n"+"Answer: " + answer["answer"]
    return faq_response
    
# Extractive QA functions

## preprocess monopoly rules

def preprocess_txt_doc(fpath):

    converter = TextConverter(remove_numeric_tables=True, valid_languages=["en"])
    doc_txt = converter.convert(file_path=fpath, meta=None)[0]
    preprocessor = PreProcessor(
        clean_empty_lines=True,
        clean_whitespace=True,
        clean_header_footer=False,
        split_by="word",
        split_length=100,
        split_respect_sentence_boundary=True,)
    docs = preprocessor.process([doc_txt])
    return docs

def start_ex_haystack(documents):
    ex_document_store = InMemoryDocumentStore()
    ex_document_store.write_documents(documents)
    retriever = TfidfRetriever(document_store=ex_document_store)
    reader = FARMReader(model_name_or_path="deepset/roberta-base-squad2", use_gpu=False)
    pipe = ExtractiveQAPipeline(reader, retriever)
    return pipe

docs = preprocess_txt_doc("monopoly_text_v1.txt")
ex_pipeline = start_ex_haystack(docs)

def predict_extract(question):
    prediction = ex_pipeline.run(question)
    possible_answers = ""
    for i,a in enumerate(prediction["answers"]):
        possible_answers =  possible_answers  +str(i) + ":" + a.answer + "\n"
    return possible_answers
 
# Gradio App section 
input_question =gr.inputs.Textbox(label="enter your monopoly question here")
response = "text"
examples = ["how much cash do we get to start with?", "at what point can I buy houses?", "what happens when I land on free parking?"]

mon_faq = gr.Interface(
            fn=predict_faq,
            inputs=input_question,
            outputs=response,
            examples=examples,
            title="Monopoly FAQ Semantic Search")

# extractive interface
mon_ex = gr.Interface(
            fn=predict_extract,
            inputs=input_question,
            outputs=response,
            examples=examples,
            title="Monopoly Extractive QA Search")
 
gradio.TabbedInterface([mon_faq,mon_ex]).launch()