from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, AutoModelForSequenceClassification, TextClassificationPipeline
import torch
import gradio as gr
from openpyxl import load_workbook
from numpy import mean

# Load tokenizers and models
tokenizer = AutoTokenizer.from_pretrained("suriya7/bart-finetuned-text-summarization")
model = AutoModelForSeq2SeqLM.from_pretrained("suriya7/bart-finetuned-text-summarization")

tokenizer_keywords = AutoTokenizer.from_pretrained("transformer3/H2-keywordextractor")
model_keywords = AutoModelForSeq2SeqLM.from_pretrained("transformer3/H2-keywordextractor")

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
new_model = AutoModelForSequenceClassification.from_pretrained('roberta-rating')
new_tokenizer = AutoTokenizer.from_pretrained('roberta-rating')

classifier = TextClassificationPipeline(model=new_model, tokenizer=new_tokenizer, device=device)

label_mapping = {1: '1/5', 2: '2/5', 3: '3/5', 4: '4/5', 5: '5/5'}

# Function to parse Excel file
def parse_xl(file_path):
    cells = []

    workbook = load_workbook(filename=file_path)
    for sheet in workbook.worksheets:
        for row in sheet.iter_rows():
            for cell in row:
                if cell.value != None:
                    cells.append(cell.value)

    return cells

# Function to evaluate reviews from Excel file
def evaluate(file):
    reviews = parse_xl(file)
    ratings = []
    text = ""
    sentiments = []

    for review in reviews:
        rating = int(classifier(review)[0]['label'].split('_')[1])
        ratings.append(rating)
        text += review
        text += " "
        
        sentiment = classifier(review)[0]['label']
        sentiment_label = "Positive" if sentiment == "LABEL_4" or sentiment == "LABEL_5" else "Negative" if sentiment == "LABEL_1" or sentiment == "LABEL_2" else "Neutral"
        sentiments.append(sentiment_label)
    
    overall_sentiment = "Positive" if sentiments.count("Positive") > sentiments.count("Negative") else "Negative" if sentiments.count("Negative") > sentiments.count("Positive") else "Neutral"
            
    inputs = tokenizer([text], max_length=1024, truncation=True, return_tensors="pt")
    summary_ids = model.generate(inputs["input_ids"], num_beams=2, min_length=10, max_length=50)
    summary = tokenizer.batch_decode(summary_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]

    # Modify the summary to third person
    summary = summary.replace("I", "He/She").replace("my", "his/her").replace("me", "him/her")

    inputs_keywords = tokenizer_keywords([text], max_length=1024, truncation=True, return_tensors="pt")
    summary_ids_keywords = model_keywords.generate(inputs_keywords["input_ids"], num_beams=2, min_length=0, max_length=100)
    keywords = tokenizer_keywords.batch_decode(summary_ids_keywords, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]   

    return round(mean(ratings), 2), summary, keywords, overall_sentiment

# Function to test a single text input
def test_area(text):
    inputs = tokenizer([text], max_length=1024, truncation=True, return_tensors="pt")
    summary_ids = model.generate(inputs["input_ids"], num_beams=2, min_length=10, max_length=50)
    summary = tokenizer.batch_decode(summary_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]

    # Modify the summary to third person
    summary = summary.replace("I", "He/She").replace("my", "his/her").replace("me", "him/her")

    inputs_keywords = tokenizer_keywords([text], max_length=1024, truncation=True, return_tensors="pt")
    summary_ids_keywords = model_keywords.generate(inputs_keywords["input_ids"], num_beams=2, min_length=0, max_length=100)
    keywords = tokenizer_keywords.batch_decode(summary_ids_keywords, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]   

    sentiment = classifier(text)[0]['label']
    sentiment_label = "Positive" if sentiment == "LABEL_4" or sentiment == "LABEL_5" else "Negative" if sentiment == "LABEL_1" or sentiment == "LABEL_2" else "Neutral"
    
    rating = int(classifier(text)[0]['label'].split('_')[1])

    return rating, summary, keywords, sentiment_label

# Main interface
main_interface = gr.Interface(
    fn=evaluate,
    inputs=gr.File(label="Reviews"),
    outputs=[gr.Textbox(label="Overall Rating"), gr.Textbox(label="Summary"), gr.Textbox(label="Keywords"), gr.Textbox(label="Overall Sentiment")],
    title='Summarize Reviews',
    description="Evaluate and summarize collection of reviews. Reviews are submitted as an Excel file, where each review is in its own cell."
)

# Testing area interface
testing_interface = gr.Interface(
    fn=test_area,
    inputs=gr.Textbox(label="Input Text"),
    outputs=[gr.Textbox(label="Rating"), gr.Textbox(label="Summary"), gr.Textbox(label="Keywords"), gr.Textbox(label="Sentiment")],
    title='Testing Area',
    description="Test the summarization, keyword extraction, sentiment analysis, and rating on custom text input."
)

# Combine interfaces into a tabbed interface with a sidebar
with gr.Blocks() as demo:
    with gr.Row():
        with gr.Column(scale=1):
            gr.Markdown("## Sidebar")
            gr.Button("Button 1")
            gr.Button("Button 2")
        with gr.Column(scale=4):
            iface = gr.TabbedInterface(
                [main_interface, testing_interface],
                ["Summarize Reviews", "Testing Area"]
            )

demo.launch(share=True)