app.py

import gradio as gr
import pandas as pd
import tiktoken

import time
from sentence_transformers import SentenceTransformer





import os
import torch

from openai.embeddings_utils import get_embedding, cosine_similarity
import os



df = pd.read_pickle('entire_data.pkl') #to load 123.pkl back to the dataframe df
embedder = SentenceTransformer('all-mpnet-base-v2')

def search(query):
  n = 15
  query_embedding = embedder.encode(query)
  df["similarity"] = df.embedding.apply(lambda x: cosine_similarity(x, query_embedding.reshape(768,-1)))

  results = (
      df.sort_values("similarity", ascending=False)
      .head(n))
  
  resultlist = []

  hlist = []
  for r in results.index:
      if results.name[r] not in hlist:
          smalldf = results.loc[results.name == results.name[r]]
          smallarr = smalldf.similarity[r].max()
          sm =smalldf.rating[r].mean()
          
          if smalldf.shape[1] > 3:
            smalldf = smalldf[:3]

          resultlist.append(
          {
            "name":results.name[r],
            "description":results.description[r],
            "relevance score": smallarr.tolist(),
            "rating": sm.tolist(),
            "relevant_reviews": [ smalldf.text[s] for s in smalldf.index]
          })
          hlist.append(results.name[r])
  return resultlist

def greet(query):
    bm25 = search(query)
    return bm25

demo = gr.Interface(fn=greet, inputs="text", outputs="json")

demo.launch()