apps/classifier.py

import json

import streamlit as st
from transformers import AutoTokenizer, RobertaForSequenceClassification, pipeline

with open("config.json") as f:
 cfg = json.loads(f.read())


@st.cache(allow_output_mutation=True, show_spinner=False)
def load_model(input_text, model_name_or_path):
 tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)
 model = RobertaForSequenceClassification.from_pretrained(model_name_or_path)

 nlp = pipeline("text-classification", model=model, tokenizer=tokenizer)
 result = nlp(input_text)
 return result


def app():
 st.title("मराठी Marathi News Classifier")

 st.markdown(
 "This demo uses the below fine-tuned models for marathi news classification:\n"
 "- [IndicNLP Marathi News Classifier](https://huggingface.co/flax-community/mr-indicnlp-classifier) fine-tuned on "
 "[IndicNLP Marathi News Dataset](https://github.com/ai4bharat/indicnlp_corpus#indicnlp-news-article-classification-dataset)\n"
 "> `IndicNLP` model predicts one of these 3 classes - `['lifestyle', 'entertainment', 'sports']`\n"
 "- [iNLTK Marathi News Classifier](https://huggingface.co/flax-community/mr-inltk-classifier) fine-tuned on "
 "[Marathi News Dataset](https://www.kaggle.com/disisbig/marathi-news-dataset)\n"
 "> `iNLTK` model predicts one of these 3 classes - `['state', 'entertainment', 'sports']`"
 )

 classifier = st.sidebar.selectbox("Select a Model", index=0, options=["IndicNLP", "iNLTK"])

 st.sidebar.markdown(
 "**IndicNLP Classes**\n"
 "- lifestyle\n"
 "- entertainment\n"
 "- sports\n"
 "\n"
 "**iNLTK Classes**\n"
 "- state\n"
 "- entertainment\n"
 "- sports"
 )

 sample_texts = [
 "रोहित शर्माने सरावाला सुरुवात करण्यापूर्वी भारतीय खेळाडूला दिला कानमंत्र, म्हणाला...",
 "जॉनी लीवर यांनी नम्रता संभेरावला दिलं खास गिफ्ट, अभिनेत्रीने व्यक्त केल्या भावना",
 "Custom",
 ]
 model_name_or_path = cfg["models"][classifier]

 text_to_classify = st.selectbox("Select a Text", options=sample_texts, index=len(sample_texts) - 1)

 if text_to_classify == "Custom":
 text_to_classify = st.text_input("Enter custom text:")

 predict_button = st.button("Predict")

 if predict_button:
 with st.spinner("Generating prediction..."):
 result = load_model(text_to_classify, model_name_or_path)

 st.markdown("## Predicted Label: `{}`".format(result[0]["label"]))
 st.markdown("## Confidence: `{}`%".format(round(result[0]["score"], 3) * 100))

 st.markdown("- - -")
 st.markdown(
 "❓ Can't figure out where to get a sample text other than the predefined ones? ❓\n"
 "\n"
 "We have provided Marathi newspaper links (section wise) below. Head over to any section of your choice, "
 "copy any headline and paste below to see if the model is predicting the respective class correctly or not?\n"
 "- [entertainment](https://maharashtratimes.com/entertainment/articlelist/19359255.cms)\n"
 "- [sports](https://maharashtratimes.com/sports/articlelist/2429056.cms)\n"
 "- [lifestyle](https://maharashtratimes.com/lifestyle-news/articlelist/2429025.cms)\n"
 "- [state](https://maharashtratimes.com/maharashtra/articlelist/2429066.cms)\n"
 "> 📒 NOTE: Both models are not trained on above headlines! Feel free to use any headline from any newspaper"
 )