homework.py

import streamlit as st
from pyparsing import empty
from mymodel import CustomBertForSequenceClassification
from transformers import BertTokenizer, Trainer

POSITIVE = 0
NEGATIVE = 1
NEUTRAL = 2

idx_target = {POSITIVE:'positive', NEGATIVE:'negative', NEUTRAL:'neutral'}

g_selected_model_type = None
g_input_text = ' '

def get_model_type(select_model, add_layers):
    if (select_model == "BERT") & (add_layers == True):
        return "klue/bert-base", "bert_layered"
    elif (select_model == "BERT") & (add_layers == False):
        return "klue/bert-base", "bert"
    elif (select_model == "RoBERTa") & (add_layers == True):
        return "klue/roberta-base", "roberta_layered"
    elif (select_model == "RoBERTa") & (add_layers == False):
        return "klue/roberta-base", "roberta"

def show_predict_result(model_type, input_text):
    target = POSITIVE
    show_response_img(target)
    show_response_text(target)

def show_response_img(target = None):
    if target == None:
        st.con3.write("")
    elif target == POSITIVE:
        st.con3.write("웃는 표정")
    elif target == NEGATIVE:
        st.con3.write("우는 표정")
    elif target == NEUTRAL:
        st.con3.write("무덤덤한 표정")

def show_response_text(target = None):
    if target == None:
        st.con4.write("")
    elif target == POSITIVE:
        st.con4.write("긍정적인 반응")
    elif target == NEGATIVE:
        st.con4.write("부정정인 반응")
    elif target == NEUTRAL:
        st.con4.write("무덤덤한 반응")
    

def show_data():
    st.write("data")

def show_api_usage():
    with st.container(border=True):
        st.write("api 사용법")

def tab1_page():
    global g_selected_model_type
    global g_input_text
    st.con1,st.con2 = st.columns([0.3,0.7])
    st.con3,st.con4 = st.columns([0.3,0.7])
    st.con5,empty1 = st.columns([0.9999,0.0001])

    with st.container():
        with st.con1:
            with st.con1.container(border=True):
                selected_model = st.selectbox("예측 모델을 선택하세요.", ["BERT", "RoBERTa"])
                add_layers = st.checkbox('Layer 추가')
                g_selected_model_type = get_model_type(selected_model, add_layers)
                show_predict_result(g_selected_model_type, g_input_text)
        with st.con2:
            with st.con2.container(border=True):
                input_text = st.text_area("input_text")
                submit_button = st.button('확인')
                if submit_button :
                    g_input_text = input_text
                    show_predict_result(g_selected_model_type, g_input_text)
    with st.con3:
        with st.con3.container(border=True):
            show_response_img()
    with st.con4:
        with st.con4.container(border=True):
            show_response_text()
    with st.con5:
        with st.con5.container(border=True):
            show_data()
    with empty1:
        empty()


def tab2_page():
    show_api_usage()


#tokenizer = BertTokenizer()
#model = CustomBertForSequenceClassification()
#trainer = Trainer()
tokenizer = BertTokenizer.from_pretrained("bert_tokenizer_layered")
model = CustomBertForSequenceClassification.from_pretrained("bert_model_layered")
trainer = Trainer().load_model("bert_trainer_layered")



st.title("Semi Project - Sentiment analysis")
#st.subheader("일상 대화를 입력하면 해당 문장이 나타내는 느낌을 표시합니다.")


tab1, tab2 = st.tabs(['동작 확인', 'API'])

with tab1:
    tab1_page()
    
with tab2:
    tab2_page()