app.py

import pandas as pd
import numpy as np
from dotenv import load_dotenv
#from keras.src.backend.jax.random import categorical

from langchain_community.document_loaders import TextLoader  #Transforms raw text into data langchain can work on
from langchain_text_splitters import CharacterTextSplitter  # Breaks the text into chunks
from langchain_openai import OpenAIEmbeddings  #Convert chunks into document embeddings
from langchain_chroma import Chroma  # Import into vector database Chroma is one type of vector database opensource

import gradio as gr

load_dotenv()
#Chroma persistance store
# import chromadb
# client = chromadb.PersistentClient()

board_games = pd.read_csv('board_games_data_clean.csv')

raw_documents = TextLoader("tagged_descriptions.txt").load()
text_splitter = CharacterTextSplitter(chunk_size=0, chunk_overlap=0, separator="\n")
documents = text_splitter.split_documents(raw_documents)
db_boardgames = Chroma.from_documents(documents, embedding=OpenAIEmbeddings())

def retrive_semantic_recomendations(
        query: str,
        min_players: str = None,
        max_players: str = None,
        game_type: str = None,
        initial_top_k: int = 50,
        final_top_k: int = 16
) -> pd.DataFrame:

    recs = db_boardgames.similarity_search(query, k=initial_top_k)
    #print(len(recs))
    game_list = [int(rec.page_content.strip('"').split()[0]) for rec in recs]
    game_recs = board_games[board_games["id"].isin(game_list)]
    #print(len(game_recs))
    if min_players!= "All":
        game_recs = game_recs[game_recs["min_players"] == min_players][:final_top_k]
    else:
        game_recs = game_recs.head(final_top_k)

    if max_players!= "All":
        game_recs = game_recs[game_recs["max_players"] == max_players][:final_top_k]
    else:
        game_recs = game_recs.head(final_top_k)

    if game_type != "All":
        game_recs = game_recs[game_recs["type"] == game_type][:final_top_k]
    else:
        game_recs = game_recs.head(final_top_k)


    return game_recs

def recommend_games(
        query: str,
        min_players: str = None,
        max_players: str = None,
        game_type: str = None,
):
    recommendations = retrive_semantic_recomendations(query, min_players, max_players, game_type)
    results = []
    for _, row in recommendations.iterrows():
        description = row["description"]
        truncated_desc_split = description.split()
        truncated_description = " ".join(truncated_desc_split[:30]) + "..."

        designers_split = row["designers_list"].split("|")

        if len(designers_split) == 2:
            designers_str = f"{designers_split[0]} {designers_split[1]}"
        elif len(designers_split) > 2:
            designers_str = f"{','.join(designers_split[:-1])} and {designers_split[-1]}"
        else:
            designers_str = row["designers_list"]

        #caption = f"{row['name']} by {designers_str} for {row['players']} players. Type: {row['type']} Rating: {row['rating']}\n {truncated_description}"
        # caption = {
        #     "name":row['name'],
        #     "designers":designers_str,
        #     "players":row['players'],
        #     "type": row['type'],
        #     "rating":row['rating'],
        #     "small_description": truncated_description,
        #     "description":   description,
        #     "thumbnail":row["thumbnail_url"]         
        # }
        caption = f"{row['name']}|{designers_str}|{row['players']}|{row['type']}|{row['rating']}|{truncated_description}|{description}"
        results.append((row["thumbnail_url"],caption)) #"cover-not-found.jpg"
        # results.append(caption) #"cover-not-found.jpg"
    #print(len(results))
    return results

# min_players = ["All"] + sorted(board_games["min_players"].unique())
# max_players = ["All"] + sorted(board_games["max_players"].unique())
game_types = ["All"] + sorted(board_games["type"].unique())

with gr.Blocks(theme = gr.themes.Glass()) as demo:
    gr.Markdown("# Semantic Boardgame recommender")
    with gr.Row():
        user_query = gr.TextArea(label="Please enter a description of a game:",
                                placeholder="e.g., A game about empire building set in space")

    with gr.Row():
        with gr.Column(scale=1, min_width=300):
            min_players_dropdown = gr.Dropdown(choices=["All","1","2","3","4","5","6","7","8"], label="Select min number of players", value="All")
            max_players_dropdown = gr.Dropdown(choices=["All","1","2","3","4","5","6","7","8","9","10"], label="Select max number of players", value="All")        
            #min_players_dropdown = gr.Dropdown(choices=min_players, label="Select min number of players", value="All")
            #max_players_dropdown = gr.Dropdown(choices=max_players, label="Select max number of players", value="All")        

        with gr.Column(scale=2, min_width=300):
            type_dropdown = gr.Dropdown(choices=game_types, label="Select a type of game", value="All")
    with gr.Row():
        submit_button = gr.Button("Find recommendations")

    with gr.Row():
        gr.Markdown("## Recommendations")
    with gr.Row():
        output = gr.Gallery(label="Recommendations", columns=8, rows=2)
        submit_button.click(fn=recommend_games, inputs=[user_query, min_players_dropdown, max_players_dropdown,type_dropdown], outputs=output)

if __name__ == "__main__":
    demo.launch()