app.py

import streamlit as st
import pandas as pd
import numpy as np
from sentence_transformers import SentenceTransformer
import faiss
import string
import re
import joblib
import time
import base64

def get_base64_of_bin_file(bin_file):
    with open(bin_file, 'rb') as f:
        data = f.read()
    return base64.b64encode(data).decode()

# Путь к вашему локальному изображению
img_file = 'fon3.jpg'  # Убедитесь, что путь правильный относительно вашего скрипта

# Преобразование изображения в base64
img_base64 = get_base64_of_bin_file(img_file)

page_bg_img = f"""
<style>
[data-testid="stAppViewContainer"] {{
background: linear-gradient(rgba(255, 255, 255, 0.5), rgba(255, 255, 255, 0.5)), url("data:image/jpeg;base64,{img_base64}");
background-size: cover;
background-repeat: no-repeat;
background-attachment: fixed;
filter: brightness(1.1); /* Adjust the brightness here */
}}
.custom-title {{
    font-size: 70px;
    font-weight: bold;
    color: #120c01;
    text-align: center;
    text-shadow: 2px 2px 4px rgba(0, 0, 0, 0.5);
}}
</style>
"""

st.markdown(page_bg_img, unsafe_allow_html=True)

# Загрузка данных
@st.cache_resource
def load_data_models():
    data = pd.read_csv('data/series_edited.csv')
    # data['description'] = data['description'].astype(str)
    combined_embeddings = np.load('embeddings/combined_embeddings_2.npy')
    index = faiss.read_index('embeddings/faiss_index_2.bin')
    embedder = SentenceTransformer('paraphrase-multilingual-mpnet-base-v2')
    lsa = joblib.load('embeddings/lsa_model.pkl')

    return data, combined_embeddings, index, embedder, lsa

data, combined_embeddings, index, embedder, lsa = load_data_models()

#Функция предобработки текста
def clean_text(text):
    text = re.sub(r'\r\n', ' ', text)
    text = text.replace('\xa0', ' ')
    clean_pattern = re.compile(r'[^a-zA-Zа-яА-ЯёЁ0-9.,!?;:\s]')
    text = clean_pattern.sub('', text)
    url_pattern = re.compile(r'http\S+|www\S+|https\S+')
    text = url_pattern.sub(r'', text)
    text = text.translate(str.maketrans('', '', string.punctuation))
    text = text.lower()
    return text

#Функция поиска подходящего по пользовательскому запросу сериала
def search_series(user_query, top_k, description_weight, actors_weight, genre_weight):
    user_query = clean_text(user_query)
    query_embedding = embedder.encode([user_query], convert_to_tensor=True).cpu().numpy()
    weighted_query_embedding = np.concatenate((query_embedding * description_weight,
                                               query_embedding * actors_weight,
                                               query_embedding * genre_weight), axis=1)
    weighted_query_embedding = lsa.transform(weighted_query_embedding)
    weighted_query_embedding = weighted_query_embedding / np.linalg.norm(weighted_query_embedding, axis=1, keepdims=True)  # Нормализация
    D, I = index.search(weighted_query_embedding, top_k)
    # results = data.iloc[I[0]].copy()
    # cosine_similarities = D[0]
    # results['cosine_similarity'] = cosine_similarities
    return I[0], D[0] 

# Инициализация весов в session_state
if 'description_weight' not in st.session_state:
    st.session_state['description_weight'] = 0.7
if 'actors_weight' not in st.session_state:
    st.session_state['actors_weight'] = 0.15
if 'genre_weight' not in st.session_state:
    st.session_state['genre_weight'] = 0.15

# Инициализация оригинальных весов для сброса
if 'original_description_weight' not in st.session_state:
    st.session_state['original_description_weight'] = 0.7
if 'original_actors_weight' not in st.session_state:
    st.session_state['original_actors_weight'] = 0.15
if 'original_genre_weight' not in st.session_state:
    st.session_state['original_genre_weight'] = 0.15

# Функция для сброса весов к первоначальным значениям
def reset_weights():
    st.session_state['description_weight'] = 0.7
    st.session_state['actors_weight'] = 0.15
    st.session_state['genre_weight'] = 0.15

# Функция для расчета пропорционального изменения весов
def update_weights():
    total_original_weight = st.session_state['original_description_weight'] + st.session_state['original_actors_weight'] + st.session_state['original_genre_weight']
    total_new_weight = st.session_state['description_weight'] + st.session_state['actors_weight'] + st.session_state['genre_weight']
    
    if total_original_weight != 0 and total_new_weight != 0:
        proportion = total_original_weight / total_new_weight
        st.session_state['description_weight'] *= proportion
        st.session_state['actors_weight'] *= proportion
        st.session_state['genre_weight'] *= proportion

# Слайдеры для настройки весов
description_weight = st.sidebar.slider("Вес описания", 0.0, 1.0, st.session_state['description_weight'], step=0.01)
actors_weight = st.sidebar.slider("Вес актеров", 0.0, 1.0, st.session_state['actors_weight'], step=0.01)
genre_weight = st.sidebar.slider("Вес жанра", 0.0, 1.0, st.session_state['genre_weight'], step=0.01)

# Обновляем значения весов в session_state
st.session_state['description_weight'] = description_weight
st.session_state['actors_weight'] = actors_weight
st.session_state['genre_weight'] = genre_weight

# Обработчик кнопки сброса весов
if st.sidebar.button("Обновить веса"):
    reset_weights()

# Обновляем веса пропорционально
update_weights()

st.markdown('<h1 class="custom-title">📽️FIND MY SHOW📽️</h1>', unsafe_allow_html=True)
# st.title('✨FIND MY SHOW✨')
st.header('Поиск сериала по описанию')
input_text = st.text_area('Введите описание сериала')
top_k = st.slider("Количество результатов", min_value=1, max_value=20, value=5)
if st.button('Поиск'):
        start_time = time.time() 
        indices, distances = search_series(input_text, top_k, description_weight, actors_weight, genre_weight)
        end_time = time.time()  
        search_time = end_time - start_time
        st.write("Результаты поиска:")
        for idx, dist in zip(indices, distances):
            results = data.iloc[idx]
            st.write("---")
            st.image(results['image_url'], width=400)
            st.write(f"**Название:** **{results['tvshow_title']}**")
            st.write(f"**Жанр:** {results['genre']}")
            if len(results['description']) > 50:
                results['description'] = ' '.join(results['description'].split()[:50]) + '...'
            st.write(f"**Описание:** {results['description']}")
            st.write(f"**Рейтинг КП :** {round(results['kinopoisk_rating'], 2)}, **Рейтинг IMDb :** {round(results['imdb'], 2)}")
            st.write(f"**Косинусное сходство:** {round(dist, 3)}")
            st.write(f'**Время поиска:** {search_time:.3f} секунд')
            st.markdown(f"[Читать далее]({results['page_url']})", unsafe_allow_html=True)