app.py

import gradio as gr
import pandas as pd
import pickle
from sentence_transformers import SentenceTransformer, util
import re

mdl_name = 'sentence-transformers/all-distilroberta-v1'
model = SentenceTransformer(mdl_name)

embedding_cache_path = "scotch_embd_distilroberta.pkl"
with open(embedding_cache_path, "rb") as fIn:
    cache_data = pickle.load(fIn)

embedding_table = cache_data["embeddings"]
reviews = cache_data["data"]
reviews['price'] = reviews.price.apply(lambda x: re.findall("\d+", x.replace(",","").replace(".00","").replace("$",""))[0]).astype('int')

def user_query_recommend(query, price_rng):
    # Embed user query
    embedding = model.encode(query)

    # Calculate similarity with all reviews
    sim_scores = util.cos_sim(embedding, embedding_table)
    #print(sim_scores.shape)

    # Recommend
    recommendations = reviews.copy()
    recommendations['sim'] = sim_scores.T
    recommendations = recommendations.sort_values('sim', ascending=False)
    if price_rng == "$0-$50":
        min_p, max_p = 0, 50
    if price_rng == "$50-$100":
        min_p, max_p = 50, 100
    if price_rng == "$100+":
        min_p, max_p = 100, 10000
    recommendations = recommendations.loc[(recommendations.price >= min_p) &
                                          (recommendations.price <= max_p),
                                          ['name', 'price', 'description']].head(5)

    return recommendations.reset_index(drop=True)
    
interface = gr.Interface(
    user_query_recommend, 
    inputs=[gr.inputs.Textbox(lines=5, label = "enter flavour profile"),
            gr.inputs.Radio(choices = ["$0-$50", "$50-$100", "$100+"], default="$0-$50", type="value", label='Price range')],
    outputs=gr.outputs.Dataframe(max_rows=1, overflow_row_behaviour="paginate", type="pandas", label="Scotch recommendations"),
    title = "Scotch Recommendation",
    description = "Looking for scotch recommendations and have some flavours in mind? \nGet recommendations at a preferred price range using semantic search :) ",
    examples=[["very sweet with lemons and oranges and marmalades", "$0-$50"], 
              ["smoky peaty earthy and spicy","$50-$100"],
              ["salty and spicy with exotic fruits", "$100+"], 
              ["fragrant nose with chocolate, toffee, pudding and caramel", "$50-$100"], 
              ],
    theme="grass",
)

interface.launch(
    enable_queue=True,
    #cache_examples=True,
)