import streamlit as st
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
import pandas as pd

st.set_page_config(page_title="Sentiment Analysis", layout="wide")

st.title("Welcome to the Sentiment Analyzer")
st.write("**Note**: All reviews must be entered in English.")


@st.cache_resource
def load_model():
  
    model_id = "Diary14/roberta-sentiment-lora"  
    
   
    tokenizer = AutoTokenizer.from_pretrained(model_id)
    model = AutoModelForSequenceClassification.from_pretrained(model_id)
    return tokenizer, model

with st.spinner("Loading model"):
    tokenizer, model = load_model()


user_input = st.text_area(
    "Enter reviews here :",
    placeholder="Example :\nThis product is amazing!\nI really don't like it....",
    height=200
)

if st.button("Analyze reviews"):
    lines = [line.strip() for line in user_input.split("\n") if line.strip() != ""]
    
    if not lines:
        st.warning("Please enter at least one review.")
    else:
        inputs = tokenizer(lines, return_tensors="pt", truncation=True, padding=True)
        
        with torch.no_grad():
            outputs = model(**inputs)
        
        logits = outputs.logits
        predictions = torch.argmax(logits, dim=-1).tolist()
        
        results = []
        for text, pred in zip(lines, predictions):
            sentiment = "Positive review" if pred == 1 else "Negative review"
            results.append({"Text/Review": text, "Sentiment": sentiment})
        
        df = pd.DataFrame(results)
        st.dataframe(df, use_container_width=True)