app.py

import streamlit as st

from torch.nn import Softmax

from model import ArxivModel, load_model
from tokenizer import get_tokenizer

from lables import num_to_classes, taxonomy

from parser import get_text_title

model = load_model()
tokenizer = get_tokenizer()

arxiv_model = ArxivModel(model, tokenizer)
softmax = Softmax(dim=1)

st.markdown("### Classification of article topics")

col1, col2 = st.columns(2)

text = ""
with col1:
    title_text = st.text_area("Write title of article", key='arxiv_title_input')

with col2:
    summary_text = st.text_area("Write summary of article (optional)", key='arxiv_sum_input')
click_button_text = st.button('Submit title and summary', key=1)

if click_button_text and summary_text.strip() != "":
    text = title_text.strip() + '\t' + summary_text.strip()
else:
    text = title_text.strip()
text = text.strip()

id_url = st.text_input("Write article's url or id", key='arxiv_id_input').strip()
click_button_url = st.button('Submit id', key=1)

if click_button_url and id_url != "":
    res = get_text_title(id_url)
    if res is not None:
        text = res[0].strip() + '\t' + res[1].strip()
        text = text.strip()
    else:
        st.markdown(f'<p style="color:#FF2D00;font-size:18px">Incorrect url or id</p>', unsafe_allow_html=True)
        text = ""

print(text)

if text != "":
    idxs = arxiv_model.get_idx_class(text, thr=0.95)[:10]

    for idx, prob in idxs:
        if taxonomy.get(num_to_classes[idx], -1) != -1:
            st.markdown("{} \t {}%".format(taxonomy.get(num_to_classes[idx], -1), round(prob * 100, 1)))
else:
    st.markdown("")