import streamlit as st import pandas as pd import numpy as np import torch from sentence_transformers import SentenceTransformer from resources.functions import recommend, find_rows_with_genres, get_mask_in_range st.markdown(f"<h1 style='text-align: center;'>Семантический поиск фильмов", unsafe_allow_html=True) df = pd.read_csv('resources/DF_FINAL.csv') genre_lists = df['ganres'].apply(lambda x: x.split(', ') if isinstance(x, str) else []) all_genres = sorted(list(set([genre for sublist in genre_lists for genre in sublist]))) st.write(f'<p style="text-align: center; font-family: Arial, sans-serif; font-size: 20px; color: white;">Количество фильмов \ для поиска {len(df)}</p>', unsafe_allow_html=True) st.header(':wrench: Панель инструментов') col1, col2, col3 = st.columns([1, 2, 2]) with col1: top_k = st.selectbox("Сколько фильмов?", options=[5, 10, 15, 20]) with col2: model_type = st.selectbox("Какой моделью пользуемся?\n ", options=['rubert-tiny2', 'msmarco-MiniLM-L-12-v3']) with col3: genres_list = st.multiselect("Какого жанра фильм?\n ", options=all_genres) if model_type == 'rubert-tiny2': model = SentenceTransformer('cointegrated/rubert-tiny2') emb = torch.load('resources/corpus_embeddings_rub.pth', map_location=torch.device('cpu')) else: model = SentenceTransformer('msmarco-MiniLM-L-12-v3') emb = torch.load('resources/corpus_embeddings_ms.pth', map_location=torch.device('cpu')) range_years = st.slider("В каком году вышел фильм?", min_value=df['year'].unique().min(), max_value=df['year'].unique().max(), value=(df['year'].unique().min(), df['year'].unique().max())) text = st.text_input('Что будем искать?') button = st.button('Начать поиск', type="primary") if text and button: if len(genres_list) == 0: mask = get_mask_in_range(df=df, range_values=range_years) else: mask1 = find_rows_with_genres(df=df, genres_list=genres_list) mask2 = get_mask_in_range(df=df, range_values=range_years) mask = mask1 & mask2 try: # emb = emb[mask] # df = df[mask] hits = recommend(model, text, emb, len(df)) st.write(f'<p style="font-family: Arial, sans-serif; font-size: 24px; color: pink; font-weight: bold;"><strong>\ {top_k} лучших рекомендаций</strong></p>', unsafe_allow_html=True) st.write('\n') mask_ind = df[mask].index.tolist() fil_hits = [hits[0][i] for i in range(len(hits[0])) if hits[0][i]['corpus_id'] in mask_ind] for i in range(top_k): col4, col5 = st.columns([3, 4]) with col4: try: st.image(df['poster'][fil_hits[i]['corpus_id']], width=300) except: st.image('https://cdnn11.img.sputnik.by/img/104126/36/1041263627_235:441:1472:1802_1920x0_80_0_0_fc2acc893b618b7c650d661fafe178b8.jpg', width=300) with col5: st.write(f"***Название:*** {df['title'][fil_hits[i]['corpus_id']]}") st.write(f"***Жанр:*** {(df['ganres'][fil_hits[i]['corpus_id']])}") st.write(f"***Описание:*** {df['description'][fil_hits[i]['corpus_id']]}") st.write(f"***Год:*** {df['year'][fil_hits[i]['corpus_id']]}") st.write(f"***Актерский состав:*** {df['cast'][fil_hits[i]['corpus_id']]}") st.write(f"***Косинусное сходство:*** {round(fil_hits[i]['score'], 2)}") st.write(f"***Ссылка на фильм : {df['url'][fil_hits[i]['corpus_id']]}***") st.markdown( "<hr style='border: 2px solid #000; margin-top: 10px; margin-bottom: 10px;'>", unsafe_allow_html=True ) except: message = '<p style="font-family: Arial, sans-serif; font-size: 24px; color: pink; font-weight: bold;"><strong>\ Подходящих вариантов нет. Измените критерии поиска или отключайте интернет и читайте\ <a href="https://huggingface.co/spaces/sakoser/rec_sys_books">книги</a>.</strong></p>' st.write(message, unsafe_allow_html=True)