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
    if price_rng == "$0-$70":
        min_p, max_p = 0, 70
    if price_rng == "$70-$150":
        min_p, max_p = 70, 150
    if price_rng == "$150+":
        min_p, max_p = 150, 10000
        
    op=recommendations\
            .groupby("name")\
            .agg({"sim":"max"})\
            .reset_index()\
            .set_axis(["name","sim"],axis='columns')

    op = pd.merge(op, 
                  recommendations[['name', 'category', 'price', 'description','description_sent','sim']],
                  how="left",on=["name",'sim'])

    op = op.sort_values('sim',ascending=False).loc[(op.price >= min_p) & (op.price <= max_p), ['name', 'price', 'description_sent']]

    return op.reset_index(drop=True).head(6)
    
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=3, 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,
)