app.py

import streamlit as st
import tensorflow as tf
import numpy as np
import pandas as pd
from transformers import *
import json
import numpy as np
import pandas as pd
from tqdm import tqdm
import os
from tensorflow.python.client import device_lib
import safetensor

PATH = './checkpoint-7500/'
SEQ_LEN = 128
tokenizer = BertTokenizer.from_pretrained('bert-base-multilingual-cased')

def create_sentiment_bert():
  # 버트 pretrained 모델 로드
  model = model = AutoModel.from_pretrained(PATH,local_files_only=True)
  # 토큰 인풋, 마스크 인풋, 세그먼트 인풋 정의
  token_inputs = tf.keras.layers.Input((SEQ_LEN,), dtype=tf.int32, name='input_word_ids')
  mask_inputs = tf.keras.layers.Input((SEQ_LEN,), dtype=tf.int32, name='input_masks')
  segment_inputs = tf.keras.layers.Input((SEQ_LEN,), dtype=tf.int32, name='input_segment')
  # 인풋이 [토큰, 마스크, 세그먼트]인 모델 정의
  bert_outputs = model([token_inputs, mask_inputs, segment_inputs])

  bert_outputs = bert_outputs[1]
  sentiment_first = tf.keras.layers.Dense(1, activation='sigmoid', kernel_initializer=tf.keras.initializers.TruncatedNormal(stddev=0.02))(bert_outputs)
  sentiment_model = tf.keras.Model([token_inputs, mask_inputs, segment_inputs], sentiment_first)

  sentiment_model.compile(loss=tf.keras.losses.BinaryCrossentropy(), metrics = ['accuracy'])
  return sentiment_model

def sentence_convert_data(data):
    global tokenizer
    tokens, masks, segments = [], [], []
    token = tokenizer.encode(data, max_length=SEQ_LEN, truncation=True, padding='max_length')
    
    num_zeros = token.count(0) 
    mask = [1]*(SEQ_LEN-num_zeros) + [0]*num_zeros 
    segment = [0]*SEQ_LEN

    tokens.append(token)
    segments.append(segment)
    masks.append(mask)

    tokens = np.array(tokens)
    masks = np.array(masks)
    segments = np.array(segments)
    return [tokens, masks, segments]

def movie_evaluation_predict(sentence):
    data_x = sentence_convert_data(sentence)
    predict = sentiment_model.predict(data_x)
    predict_value = np.ravel(predict)
    predict_answer = np.round(predict_value,0).item()

    print(predict_value)

    if predict_answer == 0:
      st.write("(부정 확률 : %.2f) 부정적인 영화 평가입니다." % (1.0-predict_value))
    elif predict_answer == 1:
      st.write("(긍정 확률 : %.2f) 긍정적인 영화 평가입니다." % predict_value)

sentiment_model = create_sentiment_bert()
movie_evaluation_predict("보던거라 계속보고있는데 전개도 느리고 주인공인 은희는 한두컷 나오면서 소극적인모습에 ")