app.py

import streamlit as st
import numpy as np
import pickle
from tensorflow.keras.models import load_model
from tensorflow.keras.preprocessing.sequence import pad_sequences
import nltk
from nltk.tokenize import word_tokenize
import ast
import pandas as pd

# Load assets
@st.cache_resource
def load_model_and_data():
    model = load_model("bigru_pos_tagger.h5")

    with open("word2idx_bigru.pkl", "rb") as f:
        word2idx = pickle.load(f)

    with open("idx2tag_bigru.pkl", "rb") as f:
        idx2tag = pickle.load(f)

    return model, word2idx, idx2tag

def pos_tag_sentence(sentence):
    tokens = sentence.split()
    token_ids = [word2idx.get(token, 1) for token in tokens]
    token_ids_padded = pad_sequences([token_ids], maxlen=MAX_LEN, padding="post")
    predictions = model.predict(token_ids_padded)[0]
    predicted_tags = [idx2tag[np.argmax(tag)] for tag in predictions][:len(tokens)]
    return list(zip(tokens, predicted_tags))

def makeTagedSentences(data):
    try:
        data = str(data)  
        data_list = ast.literal_eval(data)  
        formatted_string = " ".join([f"{word}\\{tag}" for word, tag in data_list])
        return formatted_string
    except (ValueError, SyntaxError):
        return "" 

model, word2idx, idx2tag = load_model_and_data()
MAX_LEN = 134

# Streamlit UI
st.title("Automatic Konkani POS Tagger")

user_input = st.text_area("Enter a Konkani sentence:", "")

if st.button("Tag Sentence"):
    if user_input.strip():
        if any("\\" in w for w in user_input.strip().split()):
            cleaned_input = " ".join([w.split("\\")[0] for w in user_input.strip().split()])
        else:
            cleaned_input = user_input.strip()

        tagged = pos_tag_sentence(cleaned_input)
        df = pd.DataFrame(tagged, columns=["Word", "Predicted Tag"])
        
        final_tagged_sentence = makeTagedSentences(tagged)
        st.subheader("POS Tagged Konkani Sentence:")
        st.text(final_tagged_sentence)

        st.subheader("POS Tags Table: ")
        st.table(df) 
    else:
        st.warning("Please enter some Konkani Text.")


# Footer
#st.markdown(
#    """
#    <hr style="margin-top: 2rem; margin-bottom: 1rem">
#    <div style="text-align: center; font-size: 0.9rem; color: gray;">
#        &copy; 2025 | Shrikrishna Parab | Goa University
#    </div>
#    """,
#    unsafe_allow_html=True
#)