app.py

# БЕЗ ФИЛЬТРА КАТЕГОРИЙ
# import streamlit as st
# import pandas as pd
# import numpy as np
# import torch
# from transformers import AutoTokenizer, AutoModel
# from sklearn.metrics.pairwise import pairwise_distances, cosine_similarity
# import faiss

# tokenizer = AutoTokenizer.from_pretrained("cointegrated/rubert-tiny2")
# model = AutoModel.from_pretrained("cointegrated/rubert-tiny2")

# df = pd.read_csv('data_final.csv')

# MAX_LEN = 300

# def embed_bert_cls(text, model, tokenizer):
#     t = tokenizer(text, padding=True, truncation=True, return_tensors='pt', max_length=MAX_LEN)
#     with torch.no_grad():
#         model_output = model(**{k: v.to(model.device) for k, v in t.items()})
#     embeddings = model_output.last_hidden_state[:, 0, :]
#     embeddings = torch.nn.functional.normalize(embeddings)
#     return embeddings[0].cpu().numpy()

# books_embs = np.loadtxt('vectors.txt')

# index = faiss.IndexFlatIP(books_embs.shape[1])
# index.add(books_embs)

# st.title('Приложение для рекомендации книг')

# text = st.text_input('Введите запрос:')
# top_n = st.number_input('Введите количество рекомендаций:', min_value=1, max_value=50, value=1)

# recommend_button = st.button('Найти')

# if text and recommend_button:
#     query_emb = embed_bert_cls(text, model, tokenizer)
#     D, I = index.search(query_emb.reshape(1, -1), top_n)

#     st.subheader('Топ рекомендуемых книг:')

#     for i, j in zip(I[0], D[0]):
#         col_1, col_2 = st.columns([1, 3])

#         with col_1:
#             st.image(df['image_url'][i], use_column_width=True)
#             st.write(round(j* 100, 2))
#         with col_2:
#             st.write(f'Название книги: **{df["title"][i]}**')
#             st.write(f'Автор: {df["author"][i]}')
#             st.write(f'Ссылка: {df["page_url"][i]}')
#             st.write(f'Аннотация: {df["annotation"][i]}')
# БЕЗ КЭШИРОВАНИЯ  
# import streamlit as st
# import pandas as pd
# import numpy as np
# import torch
# from transformers import AutoTokenizer, AutoModel
# from sklearn.metrics.pairwise import pairwise_distances, cosine_similarity
# import faiss

# tokenizer = AutoTokenizer.from_pretrained("cointegrated/rubert-tiny2")
# model = AutoModel.from_pretrained("cointegrated/rubert-tiny2")

# df = pd.read_csv('data_final.csv')

# MAX_LEN = 300

# def embed_bert_cls(text, model, tokenizer):
#     t = tokenizer(text, padding=True, truncation=True, return_tensors='pt', max_length=MAX_LEN)
#     with torch.no_grad():
#         model_output = model(**{k: v.to(model.device) for k, v in t.items()})
#     embeddings = model_output.last_hidden_state[:, 0, :]
#     embeddings = torch.nn.functional.normalize(embeddings)
#     return embeddings[0].cpu().numpy()

# books_embs = np.loadtxt('vectors.txt')

# index = faiss.IndexFlatIP(books_embs.shape[1])
# index.add(books_embs)

# st.title('Приложение для рекомендации книг')

# # Добавляем опциональный фильтр для выбора категории книги
# category_filter = st.selectbox('Выберите категорию книги (необязательно)', ['Все'] + list(df['category'].unique()))

# text = st.text_input('Введите запрос:')
# top_n = st.number_input('Введите количество рекомендаций:', min_value=1, max_value=50, value=1)

# recommend_button = st.button('Найти')

# if text and recommend_button:
#     query_emb = embed_bert_cls(text, model, tokenizer)
#     D, I = index.search(query_emb.reshape(1, -1), top_n)

#     st.subheader('Топ рекомендуемых книг:')

#     for i, j in zip(I[0], D[0]):
#         # Добавляем фильтрацию по выбранной категории книги, если выбрана конкретная категория
#         if category_filter == 'Все' or df['category'][i] == category_filter:
#             col_1, col_2 = st.columns([1, 3])

#             with col_1:
#                 st.image(df['image_url'][i], use_column_width=True)
#                 st.write(round(j* 100, 2))
#             with col_2:
#                 st.write(f'Название книги: **{df["title"][i]}**')
#                 st.write(f'Автор: {df["author"][i]}')
#                 st.write(f'Ссылка: {df["page_url"][i]}')
#                 st.write(f'Аннотация: {df["annotation"][i]}')
# import streamlit as st
# import pandas as pd
# import numpy as np
# import torch
# from transformers import AutoTokenizer, AutoModel
# from sklearn.metrics.pairwise import pairwise_distances, cosine_similarity
# import faiss

# tokenizer = AutoTokenizer.from_pretrained("cointegrated/rubert-tiny2")
# model = AutoModel.from_pretrained("cointegrated/rubert-tiny2")

# df = pd.read_csv('data_final.csv')

# MAX_LEN = 300

# # @st.cache(hash_funcs={tokenizers.Tokenizer: my_hash_func})
# def embed_bert_cls(text, model, tokenizer):
#     t = tokenizer(text, padding=True, truncation=True, return_tensors='pt', max_length=MAX_LEN)
#     with torch.no_grad():
#         model_output = model(**{k: v.to(model.device) for k, v in t.items()})
#     embeddings = model_output.last_hidden_state[:, 0, :]
#     embeddings = torch.nn.functional.normalize(embeddings)
#     return embeddings[0].cpu().numpy()

# @st.cache_data
# def load_faiss_index():
#     books_embs = np.loadtxt('vectors.txt')
#     index = faiss.IndexFlatIP(books_embs.shape[1])
#     index.add(books_embs)
#     return index

# st.title('Приложение для рекомендации книг')

# category_filter = st.selectbox('Выберите категорию книги (необязательно)', ['Все'] + list(df['category'].unique()))

# text = st.text_input('Введите запрос:')
# top_n = st.number_input('Введите количество рекомендаций:', min_value=1, max_value=50, value=1)

# recommend_button = st.button('Найти')

# if text and recommend_button:
#     query_emb = embed_bert_cls(text, model, tokenizer)
#     index = load_faiss_index()
#     D, I = index.search(query_emb.reshape(1, -1), top_n)

#     st.subheader('Топ рекомендуемых книг:')

#     for i, j in zip(I[0], D[0]):
#         if category_filter == 'Все' or df['category'][i] == category_filter:
#             col_1, col_2 = st.columns([1, 3])

#             with col_1:
#                 st.image(df['image_url'][i], use_column_width=True)
#                 st.write(round(j, 2))
#             with col_2:
#                 st.write(f'Название книги: **{df["title"][i]}**')
#                 st.write(f'Автор: {df["author"][i]}')
#                 st.write(f'Ссылка: {df["page_url"][i]}')
#                 st.write(f'Аннотация: {df["annotation"][i]}')

import streamlit as st
import pandas as pd
import numpy as np
import torch
from transformers import AutoTokenizer, AutoModel
import faiss

tokenizer = AutoTokenizer.from_pretrained("cointegrated/rubert-tiny2")
model = AutoModel.from_pretrained("cointegrated/rubert-tiny2")

df = pd.read_csv('data_final.csv')

MAX_LEN = 300
@st.cache_data
def load_faiss_index():
    books_embs = np.loadtxt('vectors.txt')
    index = faiss.IndexFlatIP(books_embs.shape[1])
    index.add(books_embs)
    return index

def embed_bert_cls(text, model, tokenizer):
    t = tokenizer(text, padding=True, truncation=True, return_tensors='pt', max_length=MAX_LEN)
    with torch.no_grad():
        model_output = model(**{k: v.to(model.device) for k, v in t.items()})
    embeddings = model_output.last_hidden_state[:, 0, :]
    embeddings = torch.nn.functional.normalize(embeddings)
    return embeddings[0].cpu().numpy()

@st.cache_data()
def get_recommendations(query_emb, top_n):
    index = load_faiss_index()
    D, I = index.search(query_emb.reshape(1, -1), top_n)
    recommendations = []
    for i, j in zip(I[0], D[0]):
        recommendation = {
            'image_url': df['image_url'][i],
            'title': df['title'][i],
            'author': df['author'][i],
            'page_url': df['page_url'][i],
            'annotation': df['annotation'][i],
            'category': df['category'][i],
            'similarity_score': round(j, 2)
        }
        recommendations.append(recommendation)
    return recommendations

st.title('Приложение для рекомендации книг')

category_filter = st.selectbox('Выберите категорию книги (необязательно)', ['Все'] + list(df['category'].unique()))

text = st.text_input('Введите запрос:')
top_n = st.number_input('Введите количество рекомендаций:', min_value=1, max_value=50, value=1)

recommend_button = st.button('Найти')

if text and recommend_button:
    query_emb = embed_bert_cls(text, model, tokenizer)
    recommendations = get_recommendations(query_emb, top_n)

    if not recommendations:  # Если рекомендации не найдены
        st.write('По вашему запросу ничего не найдено.')
    else:
        st.subheader('Топ рекомендуемых книг:')
        for recommendation in recommendations:
            if category_filter == 'Все' or recommendation['category'] == category_filter:
                col_1, col_2 = st.columns([1, 3])
                with col_1:
                    st.image(recommendation['image_url'], use_column_width=True)
                    st.write(recommendation['similarity_score'])
                with col_2:
                    st.write(f'Название книги: **{recommendation["title"]}**')
                    st.write(f'Автор: {recommendation["author"]}')
                    st.write(f'Ссылка: {recommendation["page_url"]}')
                    st.write(f'Аннотация: {recommendation["annotation"]}')
                    st.write(f'Категория: {recommendation["category"]}')