import zipfile
import os
import gradio as gr
import tensorflow as tf
from transformers import BertTokenizer, TFBertForSequenceClassification
import numpy as np

# Unzip model.zip if not already extracted
if not os.path.exists("model"):
    with zipfile.ZipFile("model.zip", 'r') as zip_ref:
        zip_ref.extractall("model")

# Correct Model Path
MODEL_PATH = "model"

# Load model and tokenizer
model = TFBertForSequenceClassification.from_pretrained(MODEL_PATH)
tokenizer = BertTokenizer.from_pretrained(MODEL_PATH)

# Prediction function
def predict_value(text, reason, threshold=0.7):
    combined_text = text + " [SEP] " + reason
    encoding = tokenizer(combined_text, padding="max_length", truncation=True, max_length=128, return_tensors="tf")

    logits = model.predict(dict(encoding)).logits
    probs = tf.nn.softmax(logits, axis=1).numpy()
    
    prediction = 1 if probs[:, 1] > threshold else 0
    confidence = probs[:, 1][0]

    if prediction == 1:
        result = "✅ Valuable Feedback"
    else:
        result = "❌ Not Valuable Feedback"
        
    return result, f"Confidence Score: {confidence:.2f}"

# Gradio UI
with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown(
        """
        # 🚀 Text & Reason Evaluator
        Analyze if the provided text and reason are valuable!
        """
    )

    with gr.Row():
        text_input = gr.Textbox(label="📝 Enter the Text")
        reason_input = gr.Textbox(label="💡 Enter the Reason")

    predict_button = gr.Button("🔍 Predict")

    output_result = gr.Textbox(label="Result")
    output_confidence = gr.Textbox(label="Confidence Score")

    predict_button.click(
        predict_value,
        inputs=[text_input, reason_input],
        outputs=[output_result, output_confidence],
    )

demo.launch()