pages/TaskOne.py

import streamlit as st
import torch
import torch.nn as nn
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.linear_model import LogisticRegression
import pickle
import time
from dopset import (LSTMClassifier, preprocess_single_string)
import numpy as np
import pandas as pd
from transformers import DistilBertModel, DistilBertTokenizer

loaded = pickle.load(open('BertWeight/log.pkl', "rb"))
model_BERT = DistilBertModel.from_pretrained("BertWeight/pt_save_pretrained")
tokenizer = DistilBertTokenizer.from_pretrained('distilbert-base-uncased')

st.title('Отзывы')

@st.cache_resource

def load_model_and_vectorizer():
    
    loaded_vectorizer = pickle.load(open('tfidf/tfidf_vectorizer.pkl', 'rb'))
    loaded_model = pickle.load(open('tfidf/logistic_regression_model.pkl', 'rb'))
    return loaded_vectorizer, loaded_model

def logreg(text):
    weight_vect, weight_model = load_model_and_vectorizer()

    start_time = time.time()
    input_ids = weight_vect.transform([text])
    output = weight_model.predict(input_ids)
    end_time = time.time()

    if output == 1:
         return "положительный", round(end_time - start_time, 5)
    else:
        return "Негативный", round(end_time - start_time, 5)
    
def lstm_(text):
    EMBEDDING_DIM = 64
    HIDDEN_DIM = 16
    DEVICE = 'cpu'

    model = LSTMClassifier(embedding_dim=EMBEDDING_DIM, hidden_size=HIDDEN_DIM).to(DEVICE)
    model.load_state_dict(torch.load('lstm/lstm_weights.pt', map_location=DEVICE))
    start_time = time.time()
    pred = model(preprocess_single_string(text, seq_len=128).unsqueeze(0).to(DEVICE)).sigmoid().round().item()
    end_time = time.time()
    if pred == 1:
         return "положительный", round(end_time - start_time, 5)
    else:
        return "Негативный", round(end_time - start_time, 5)
    
def bert_(text, model, loaded_model):
    start_time = time.time()
    tokenized_text = tokenizer.encode(text, add_special_tokens=True, truncation=True, max_length=64)
    input_ids = torch.tensor(tokenized_text).unsqueeze(0)
    with torch.no_grad():
        outputs = model(input_ids)
        vectors = outputs[0][:,0,:].detach().cpu().numpy()
    
    prediction = loaded_model.predict(vectors).item()
    end_time = time.time()
    if prediction == 1:
        return "положительный", round(end_time - start_time, 5)
    else:
        return "Негативный", round(end_time - start_time, 5)


table_f1 = {'Model': ['Tf-IDF + logreg', 'LSTM', 'Bert'],
            'F1-score':['0.91', '0.94', '0.74']}

df = pd.DataFrame(table_f1)

text = st.text_input("Напишите отзыв")

if text:

    rate, time_ = logreg(text)
    st.markdown('### Tf-IDF + logreg')
    st.write('Отзыв:', rate)
    st.write('Время:', time_)

    rate_lstm, time_lstm = lstm_(text)
    st.markdown('### LSTM')
    st.write('Отзыв:', rate_lstm)
    st.write('Время:', time_lstm)

    rate_bert, time_bert = bert_(text, model_BERT, loaded)
    st.markdown('### BERT')
    st.write('Отзыв:', rate_bert)
    st.write('Время:', time_bert)
    st.table(df)