app.py

import streamlit as st
import json
import time
import faiss
from sentence_transformers import SentenceTransformer
from sentence_transformers.cross_encoder import CrossEncoder


class DocumentSearch:
    '''
        This class is dedicated to
        perform semantic document search
        based on previously trained:
        faiss: index
        sbert: encoder
        sbert: cross_encoder
    '''
    def __init__(self, labels_path: str, encoder_path: str,
                 index_path: str, cross_encoder_path: str):
        # loading docs and corresponding urls
        with open(labels_path, 'r') as json_file:
            self.docs = json.load(json_file)
        # loading sbert encoder model
        self.encoder = SentenceTransformer(encoder_path)
        # loading faiss index
        self.index = faiss.read_index(index_path)
        # loading sbert cross_encoder
        self.cross_encoder = CrossEncoder(cross_encoder_path)

    def search(self, query: str, k: int) -> list:
        # get vector representation of text query
        query_vector = self.encoder.encode([query])
        # perform search via faiss FlatIP index
        _, indeces = self.index.search(query_vector, k*10)
        # get answers by index
        answers = [self.docs[i] for i in indeces[0]]
        # prepare inputs for cross encoder
        model_inputs = [[query, pairs[0]] for pairs in answers]
        urls = [pairs[1] for pairs in answers]
        # get similarity score between query and documents
        scores = self.cross_encoder.predict(model_inputs, batch_size=1)
        # compose results into list of dicts
        results = [{'doc': doc[1], 'url': url, 'score': score} for doc, url, score in zip(model_inputs, urls, scores)]
        # return results sorteed by similarity scores
        return sorted(results, key=lambda x: x['score'], reverse=True)[:k]


enc_path = "ivan-savchuk/msmarco-distilbert-dot-v5-tuned-full-v1"
idx_path = "idx_vectors.index"
cross_enc_path = "ivan-savchuk/cross-encoder-ms-marco-MiniLM-L-12-v2-tuned_mediqa-v1"
docs_path = "docs.json"
# get instance of DocumentSearch class
surfer = DocumentSearch(
    labels_path=docs_path,
    encoder_path=enc_path,
    index_path=idx_path,
    cross_encoder_path=cross_enc_path
)


if __name__ == "__main__":
    # streamlit part starts here with title
    st.title('Medical Search')
    # here we have input space
    query = st.text_input("Enter any query about our data",
                          placeholder="Type query here...")
    # on submit we execute search
    if(st.button("Search")):
        # set start time
        stt = time.time()
        # retrieve top 5 documents
        results = surfer.search(query, k=5)
        # set endtime
        ent = time.time()
        # measure resulting time
        elapsed_time = round(ent - stt, 2)

        # define container for answers
        with st.container():
            # show which query was entered, and what was searching time
            st.write(f"**Results Related to:** {query} ({elapsed_time} sec.)")
            # then we use loop to show results
            for i, answer in enumerate(results):
                # answer starts with header
                st.subheader(f"Answer {i+1}")
                # cropped answer
                doc = answer["doc"][:150] + "..."
                # and url to the full answer
                url = answer["url"]
                # then we display it
                st.markdown(f"{doc}\n[**Read More**]({url})\n")