# Dependencies:
# gradio==3.3.1
# transformers==4.27.1
# torch==2.0.1
# pymupdf==1.21.1

import gradio as gr
from transformers import pipeline
import fitz  # PyMuPDF

# Load a summarization model from Hugging Face
summarizer = pipeline("summarization")

def extract_text_from_pdf(pdf_file):
    text = ""
    try:
        if isinstance(pdf_file, bytes):
            document = fitz.open(stream=pdf_file, filetype="pdf")
        else:
            document = fitz.open(stream=pdf_file.read(), filetype="pdf")
        for page_num in range(len(document)):
            page = document.load_page(page_num)
            text += page.get_text()
    except Exception as e:
        return str(e)
    return text

def evaluate_text_against_rubric(rubric_text, text):
    # Split rubric into criteria
    criteria = [criterion.strip() for criterion in rubric_text.split('\n') if criterion.strip()]
    
    if not criteria:
        return "No valid criteria found in the rubric."
    
    if not text:
        return "No text provided for evaluation."
    
    evaluations = {}
    for i, criterion in enumerate(criteria):
        try:
            summary = summarizer(text, max_length=50, min_length=25, do_sample=False)[0]['summary_text']
            evaluations[f'Criteria {i+1}'] = {
                "Criterion": criterion,
                "Score": 3,  # Dummy score for now
                "Comment": f"Evaluation based on criterion: {criterion}",
                "Example": summary
            }
        except Exception as e:
            evaluations[f'Criteria {i+1}'] = {
                "Criterion": criterion,
                "Score": 0,
                "Comment": f"Error during evaluation: {str(e)}",
                "Example": ""
            }
    
    return evaluations

def evaluate(rubric_pdf, rubric_text, text):
    rubric = ""
    if rubric_pdf is not None:
        rubric = extract_text_from_pdf(rubric_pdf)
    elif rubric_text:
        rubric = rubric_text
    
    if not rubric:
        return "No rubric provided."
    
    if not text:
        return "No text provided for evaluation."
    
    evaluation = evaluate_text_against_rubric(rubric, text)
    
    if isinstance(evaluation, str):  # If it's an error message
        return evaluation
    
    evaluation_text = ""
    for criterion, details in evaluation.items():
        evaluation_text += f"{criterion}:\n"
        evaluation_text += f"  Criterion: {details['Criterion']}\n"
        evaluation_text += f"  Score: {details['Score']}\n"
        evaluation_text += f"  Comment: {details['Comment']}\n"
        evaluation_text += f"  Example: {details['Example']}\n\n"
    
    return evaluation_text

# Create Gradio interface
with gr.Blocks() as interface:
    gr.Markdown("# PDF Text Evaluator")
    gr.Markdown("Upload a rubric as a PDF or paste the rubric text, then paste text for evaluation.")
    
    rubric_pdf_input = gr.File(label="Upload Rubric PDF (optional)", type="binary")
    rubric_text_input = gr.Textbox(lines=10, placeholder="Or enter your rubric text here...", label="Rubric Text (optional)")
    text_input = gr.Textbox(lines=10, placeholder="Paste the text to be evaluated here...", label="Text to Evaluate")
    
    evaluate_button = gr.Button("Evaluate")
    
    output = gr.Textbox(label="Evaluation Results")
    
    def evaluate_button_clicked(rubric_pdf, rubric_text, text):
        return evaluate(rubric_pdf, rubric_text, text)
    
    evaluate_button.click(evaluate_button_clicked, inputs=[rubric_pdf_input, rubric_text_input, text_input], outputs=output)

# Launch the interface
interface.launch()