from llmgaurdrails.custom_models.groundedness_checker.pdf_data_chunker import process_pdf
from llmgaurdrails.custom_models.groundedness_checker.llm_based_qa_generator import  LLMBasedQAGenerator
from llmgaurdrails.custom_models.groundedness_checker.grounding_classifier import GroundingTrainer
from llmgaurdrails.custom_models.groundedness_checker.simple_qa_generator import SimpleQAGenerator
from llmgaurdrails.custom_models.groundedness_checker.evaluate_groundedness_model import evaluate,get_eval_data

if __name__ == "__main__":
    
    # Replace with your PDF Files
    trainning_pdf_paths = ["D:\Sasidhar\Projects\cba\data\CreditCard.pdf" ,
                 "D:\Sasidhar\Projects\cba\data\home_insurance_pds.pdf"]
    
    eval_pdf_paths = ["D:\Sasidhar\Projects\llm_gaurdrails\llmgaurdrails\data\PrivateBanking.pdf"]
    
    all_chunks = []

    for path in trainning_pdf_paths:
        chunks = process_pdf(trainning_pdf_paths[0])
        all_chunks.append(chunks)

    chunks_flattened = [x for xs in all_chunks for x in xs]
    
    # generate qa dataset 
    qa_generator = LLMBasedQAGenerator()

    dataset = qa_generator.generate_dataset(chunks_flattened,persist_dataset=True)

    trainer = GroundingTrainer()
    trainer.train(dataset)

    eval_dataset = get_eval_data(eval_pdf_paths=eval_pdf_paths)
    evaluate(dataset)

    # Result on test dataset - This is wrong as these numbers were obtained on a trained dataset by mistake  . Will fix before the presentation.
    
    # Accuracy: 0.8952380952380953
    # Precision: 0.8738738738738738
    # Recall: 0.9238095238095239
    # F1 Score: 0.8981481481481481