import streamlit as st
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
import re

tokenizer = AutoTokenizer.from_pretrained("nebiyu29/fintunned-v2-roberta_GA")
model = AutoModelForSequenceClassification.from_pretrained("nebiyu29/fintunned-v2-roberta_GA")

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = model.to(device)

# Define a function to split a text into segments of 512 tokens
def split_text(text):
    #remove unnessary charachters
    text=re.sub(r'[^a-zA-Z\s]','',text)
    # Tokenize the text
    tokens = tokenizer.tokenize(text)
    # Initialize an empty list for segments
    segments = []
    # Initialize an empty list for current segment
    current_segment = []
    # Initialize a counter for tokens
    token_count = 0
    # Loop through the tokens
    for token in tokens:
        # Add the token to the current segment
        current_segment.append(token)
        # Increment the token count
        token_count += 1
        # If the token count reaches 512 or the end of the text, add the current segment to the segments list
        if token_count == 512 or token == tokens[-1]:
            # Convert the current segment to a string and add it to the segments list
            segments.append(tokenizer.convert_tokens_to_string(current_segment))
            # Reset the current segment and the token count
            current_segment = []
            token_count = 0
    # Return the segments list
    return segments

# Define a function to extract predictions from model output (adjust as needed)
def extract_predictions(outputs):
    # Assuming outputs contain logits and labels (adapt based on your model's output format)
    logits = outputs.logits
    probs = logits.softmax(dim=1)
    preds = torch.argmax(probs, dim=1)
    return probs, preds  # Return all probabilities and predicted labels

# a function that classifies text
def classify_text(text):
    # Split text into segments using split_text
    segments = split_text(text)

    # Initialize empty list for predictions
    predictions = []

    # Loop through segments, process, and store predictions
    for segment in segments:
        inputs = tokenizer([segment], padding=True, return_tensors="pt")
        input_ids = inputs["input_ids"].to(device)
        attention_mask = inputs["attention_mask"].to(device)

        with torch.no_grad():
            outputs = model(input_ids, attention_mask=attention_mask)

        # Extract predictions for each segment
        probs, preds = extract_predictions(outputs)  # Define this function based on your model's output
        pred_label = model.config.id2label[preds[0].item()]

        # Append predictions for this segment
        predictions.append({
            "segment_text": segment,
            "label": pred_label,  # Assuming single label prediction
            "probability": probs[0][preds[0]].item()  # Access probability for the predicted label
        })

    return predictions

# Streamlit app
st.title("Text Classification Demo")
st.write("Enter some text, and the model will classify it.")

text_input = st.text_input("Text Input")
if st.button("Classify"):
    predictions = classify_text(text_input)
    for prediction in predictions:
        st.write(f"Segment Text: {prediction['segment_text']}")
        st.write(f"Label: {prediction['label']}")
        st.write(f"Probability: {prediction['probability']}")