app.py

import streamlit as st
from streamlit_tags import st_tags, st_tags_sidebar
from keytotext import pipeline
from PIL import Image

import json
from sentence_transformers import SentenceTransformer, CrossEncoder, util
import gzip
import os
import torch

############
## Main page
############

st.write("# Code for Query Expansion")

st.markdown("***Idea is to build a model which will take query as inputs and generate expansion information as outputs.***")
image = Image.open('top.png')
st.image(image)

st.sidebar.write("# Parameter Selection")
maxtags_sidebar = st.sidebar.slider('Number of query allowed?', 1, 10, 1, key='ehikwegrjifbwreuk')
user_query = st_tags(
    label='# Enter Query:',
    text='Press enter to add more',
    value=['Mother'],
    suggestions=['five', 'six', 'seven', 'eight', 'nine', 'three', 'eleven', 'ten', 'four'],
    maxtags=maxtags_sidebar,
    key="aljnf")

# Add selectbox in streamlit
option1 = st.sidebar.selectbox(
     'Which transformers model would you like to be selected?',
     ('multi-qa-MiniLM-L6-cos-v1'))

option2 = st.sidebar.selectbox(
     'Which corss-encoder model would you like to be selected?',
     ('cross-encoder/ms-marco-MiniLM-L-6-v2'))

if not torch.cuda.is_available():
    print("Warning: No GPU found. Please add GPU to your notebook")


#We use the Bi-Encoder to encode all passages, so that we can use it with sematic search
bi_encoder = SentenceTransformer(option1)
bi_encoder.max_seq_length = 256    #Truncate long passages to 256 tokens
top_k = 32                          #Number of passages we want to retrieve with the bi-encoder

#The bi-encoder will retrieve 100 documents. We use a cross-encoder, to re-rank the results list to improve the quality
cross_encoder = CrossEncoder(option2)

# As dataset, we use Simple English Wikipedia. Compared to the full English wikipedia, it has only
# about 170k articles. We split these articles into paragraphs and encode them with the bi-encoder

etsy_filepath = '000000000001.json'

#if not os.path.exists(wikipedia_filepath):
#    util.http_get('http://sbert.net/datasets/simplewiki-2020-11-01.jsonl.gz', wikipedia_filepath)

passages = []
'''
with gzip.open(wikipedia_filepath, 'rt', encoding='utf8') as fIn:
    for line in fIn:
        data = json.loads(line.strip())

        #Add all paragraphs
        #passages.extend(data['paragraphs'])

        #Only add the first paragraph
        passages.append(data['paragraphs'][0])
'''

with open(etsy_filepath, 'r') as EtsyJson:
  for line in EtsyJson:
    data = json.loads(line.strip())
    #passages.append(data['query'])
    passages.append(data['title'])


print("Passages:", len(passages))

# We encode all passages into our vector space. This takes about 5 minutes (depends on your GPU speed)
corpus_embeddings = bi_encoder.encode(passages, convert_to_tensor=True, show_progress_bar=True)

# This function will search all wikipedia articles for passages that
# answer the query
def search(query):
    print("Input question:", query)

    ##### BM25 search (lexical search) #####
    #bm25_scores = bm25.get_scores(bm25_tokenizer(query))
    #top_n = np.argpartition(bm25_scores, -5)[-5:]
    #bm25_hits = [{'corpus_id': idx, 'score': bm25_scores[idx]} for idx in top_n]
    #bm25_hits = sorted(bm25_hits, key=lambda x: x['score'], reverse=True)

    #print("Top-10 lexical search (BM25) hits")
    #for hit in bm25_hits[0:10]:
    #    print("\t{:.3f}\t{}".format(hit['score'], passages[hit['corpus_id']].replace("\n", " ")))

    ##### Sematic Search #####
    # Encode the query using the bi-encoder and find potentially relevant passages
    query_embedding = bi_encoder.encode(query, convert_to_tensor=True)
    query_embedding = query_embedding.cuda()
    hits = util.semantic_search(query_embedding, corpus_embeddings, top_k=top_k)
    hits = hits[0]  # Get the hits for the first query

    ##### Re-Ranking #####
    # Now, score all retrieved passages with the cross_encoder
    cross_inp = [[query, passages[hit['corpus_id']]] for hit in hits]
    cross_scores = cross_encoder.predict(cross_inp)

    # Sort results by the cross-encoder scores
    for idx in range(len(cross_scores)):
        hits[idx]['cross-score'] = cross_scores[idx]

    # Output of top-10 hits from bi-encoder
    print("\n-------------------------\n")
    print("Top-10 Bi-Encoder Retrieval hits")
    hits = sorted(hits, key=lambda x: x['score'], reverse=True)
    for hit in hits[0:10]:
        print("\t{:.3f}\t{}".format(hit['score'], passages[hit['corpus_id']].replace("\n", " ")))

    # Output of top-10 hits from re-ranker
    print("\n-------------------------\n")
    print("Top-10 Cross-Encoder Re-ranker hits")
    hits = sorted(hits, key=lambda x: x['cross-score'], reverse=True)
    for hit in hits[0:10]:
        print("\t{:.3f}\t{}".format(hit['cross-score'], passages[hit['corpus_id']].replace("\n", " ")))

st.write("## Results:")
if st.button('Generate Sentence'):
    out = search(query = user_query)