app.py

import gradio as gr
import logging
from data_handler import download_nltk_resources
from analysis_runner import run_analysis
from visualization_handler import create_visualization_components
from ui.dataset_input import create_dataset_input, load_example_dataset
from ui.analysis_screen import create_analysis_screen

# Set up logging
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
logger = logging.getLogger('gradio_app')

# Import the process_analysis_request function
# Try to use the improved version if available, otherwise use original
try:
    from improved_analysis_handler import process_analysis_request

    logger.info("Using improved analysis handler")
except ImportError:
    logger.info("Using original analysis handler")
    from ui.analysis_screen import process_analysis_request


def create_app():
    """
    Create a streamlined Gradio app for dataset input and analysis.

    Returns:
        gr.Blocks: The Gradio application
    """
    with gr.Blocks(title="LLM Response Comparator") as app:
        # Application state to share data between tabs
        dataset_state = gr.State({})
        analysis_results_state = gr.State({})

        # Dataset Input Tab
        with gr.Tab("Dataset Input"):
            dataset_inputs, example_dropdown, load_example_btn, create_btn, prompt, response1, model1, response2, model2 = create_dataset_input()

            # Add status indicator to show when dataset is created
            dataset_status = gr.Markdown("*No dataset loaded*")

            # Load example dataset
            load_example_btn.click(
                fn=load_example_dataset,
                inputs=[example_dropdown],
                outputs=[prompt, response1, model1, response2, model2]  # Update all field values
            )

            # Save dataset to state and update status
            def create_dataset(p, r1, m1, r2, m2):
                if not p or not r1 or not r2:
                    return {}, "❌ **Error:** Please fill in at least the prompt and both responses"

                dataset = {
                    "entries": [
                        {"prompt": p, "response": r1, "model": m1 or "Model 1"},
                        {"prompt": p, "response": r2, "model": m2 or "Model 2"}
                    ]
                }
                return dataset, "✅ **Dataset created successfully!** You can now go to the Analysis tab"

            create_btn.click(
                fn=create_dataset,
                inputs=[prompt, response1, model1, response2, model2],
                outputs=[dataset_state, dataset_status]
            )

        # Analysis Tab
        with gr.Tab("Analysis"):
            # Create analysis screen
            analysis_components = create_analysis_screen()
            analysis_options = analysis_components[0]
            analysis_params = analysis_components[1]
            run_analysis_btn = analysis_components[2]
            analysis_output = analysis_components[3]
            ngram_n = analysis_components[4]
            topic_count = analysis_components[5]

            # Create visualization components
            visualization_components = create_visualization_components()

            # Connect the run button to the analysis function
            run_analysis_btn.click(
                fn=run_analysis,
                inputs=[dataset_state, analysis_options, ngram_n, ngram_top, topic_count],  # Make sure ngram_top is included
                outputs=visualization_components
            )

    return app


if __name__ == "__main__":
    # Download required NLTK resources before launching the app
    download_nltk_resources()

    logger.info("Starting LLM Response Comparator application")
    logger.info("===== Application Startup =====")

    # Create and launch the application
    app = create_app()
    app.launch()