app.py

import gradio as gr
import json

# Import your modules here
from Agents.togetherAIAgent import generate_article_from_query
from Agents.wikiAgent import get_wiki_data
from Agents.rankerAgent import rankerAgent
from Query_Modification.QueryModification import query_Modifier, getKeywords
from Ranking.RRF.RRF_implementation import reciprocal_rank_fusion_three, reciprocal_rank_fusion_six
from Retrieval.tf_idf import tf_idf_pipeline
from Retrieval.bm25 import bm25_pipeline
from Retrieval.vision import vision_pipeline
from Retrieval.openSource import open_source_pipeline
from Baseline.boolean import boolean_pipeline
from AnswerGeneration.getAnswer import generate_answer_withContext, generate_answer_zeroShot

# Load miniWikiCollection
miniWikiCollection = json.load(open('Datasets/mini_wiki_collection.json', 'r'))
miniWikiCollectionDict = {wiki['wikipedia_id']: " ".join(wiki['text']) for wiki in miniWikiCollection}

def process_query(query):
    # Query modification
    modified_query = query_Modifier(query)

    # Context Generation
    article = generate_article_from_query(query)

    # Keyword Extraction and getting context from Wiki
    keywords = getKeywords(query)
    wiki_data = get_wiki_data(keywords)

    # Retrieve rankings
    boolean_ranking = boolean_pipeline(query)
    tf_idf_ranking = tf_idf_pipeline(query)
    bm25_ranking = bm25_pipeline(query)
    vision_ranking = vision_pipeline(query)
    open_source_ranking = open_source_pipeline(query)

    # Modified queries
    boolean_ranking_modified = boolean_pipeline(modified_query)
    tf_idf_ranking_modified = tf_idf_pipeline(modified_query)
    bm25_ranking_modified = bm25_pipeline(modified_query)
    vision_ranking_modified = vision_pipeline(modified_query)
    open_source_ranking_modified = open_source_pipeline(modified_query)

    # RRF rankings
    tf_idf_bm25_open_RRF_Ranking = reciprocal_rank_fusion_three(tf_idf_ranking, bm25_ranking, open_source_ranking)
    tf_idf_bm25_open_RRF_Ranking_modified = reciprocal_rank_fusion_three(tf_idf_ranking_modified, bm25_ranking_modified, open_source_ranking_modified)
    tf_idf_bm25_open_RRF_Ranking_combined = reciprocal_rank_fusion_six(
        tf_idf_ranking, bm25_ranking, open_source_ranking,
        tf_idf_ranking_modified, bm25_ranking_modified, open_source_ranking_modified
    )




    agent1_context = wiki_data[0]
    agent2_context = article

    boolean_context = miniWikiCollectionDict[boolean_ranking[0]]
    tf_idf_context = miniWikiCollectionDict[tf_idf_ranking[0]]
    bm25_context = miniWikiCollectionDict[str(bm25_ranking[0])]
    vision_context = miniWikiCollectionDict[vision_ranking[0]]
    open_source_context = miniWikiCollectionDict[open_source_ranking[0]]

    boolean_context_modified = miniWikiCollectionDict[boolean_ranking_modified[0]]
    tf_idf_context_modified = miniWikiCollectionDict[tf_idf_ranking_modified[0]]
    bm25_context_modified = miniWikiCollectionDict[str(bm25_ranking_modified[0])]
    vision_context_modified = miniWikiCollectionDict[vision_ranking_modified[0]]
    open_source_context_modified = miniWikiCollectionDict[open_source_ranking_modified[0]]

    tf_idf_bm25_open_RRF_Ranking_context = miniWikiCollectionDict[tf_idf_bm25_open_RRF_Ranking[0]]
    tf_idf_bm25_open_RRF_Ranking_modified_context = miniWikiCollectionDict[tf_idf_bm25_open_RRF_Ranking_modified[0]]
    tf_idf_bm25_open_RRF_Ranking_combined_context = miniWikiCollectionDict[tf_idf_bm25_open_RRF_Ranking_combined[0]]



    #Generating answers

    agent1_answer = generate_answer_withContext(query, agent1_context)
    agent2_answer = generate_answer_withContext(query, agent2_context)

    boolean_answer = generate_answer_withContext(query, boolean_context)
    tf_idf_answer = generate_answer_withContext(query, tf_idf_context)
    bm25_answer = generate_answer_withContext(query, bm25_context)
    vision_answer = generate_answer_withContext(query, vision_context)
    open_source_answer = generate_answer_withContext(query, open_source_context)

    boolean_answer_modified = generate_answer_withContext(modified_query, boolean_context_modified)
    tf_idf_answer_modified = generate_answer_withContext(modified_query, tf_idf_context_modified)
    bm25_answer_modified = generate_answer_withContext(modified_query, bm25_context_modified)
    vision_answer_modified = generate_answer_withContext(modified_query, vision_context_modified)
    open_source_answer_modified = generate_answer_withContext(modified_query, open_source_context_modified)

    tf_idf_bm25_open_RRF_Ranking_answer = generate_answer_withContext(query, tf_idf_bm25_open_RRF_Ranking_context)
    tf_idf_bm25_open_RRF_Ranking_modified_answer = generate_answer_withContext(modified_query, tf_idf_bm25_open_RRF_Ranking_modified_context)
    tf_idf_bm25_open_RRF_Ranking_combined_answer = generate_answer_withContext(query, tf_idf_bm25_open_RRF_Ranking_combined_context)

    zeroShot = generate_answer_zeroShot(query)


    # Ranking the best answer
    rankerAgentInput = {
        "query": query,
        "agent1": agent1_answer,
        "agent2": agent2_answer,
        "boolean": boolean_answer,
        "tf_idf": tf_idf_answer,
        "bm25": bm25_answer,
        "vision": vision_answer,
        "open_source": open_source_answer,
        "boolean_modified": boolean_answer_modified,
        "tf_idf_modified": tf_idf_answer_modified,
        "bm25_modified": bm25_answer_modified,
        "vision_modified": vision_answer_modified,
        "open_source_modified": open_source_answer_modified,
        "tf_idf_bm25_open_RRF_Ranking": tf_idf_bm25_open_RRF_Ranking_answer,
        "tf_idf_bm25_open_RRF_Ranking_modified": tf_idf_bm25_open_RRF_Ranking_modified_answer,
        "tf_idf_bm25_open_RRF_Ranking_combined": tf_idf_bm25_open_RRF_Ranking_combined_answer,
        "zeroShot": zeroShot
    }

    best_model, best_answer = rankerAgent(rankerAgentInput)



    return (
        best_model,
        best_answer,
        agent1_answer, agent1_context,
        agent2_answer, agent2_context,
        boolean_answer, boolean_context,
        tf_idf_answer, tf_idf_context,
        bm25_answer, bm25_context,
        vision_answer, vision_context,
        open_source_answer, open_source_context,
        boolean_answer_modified, boolean_context_modified,
        tf_idf_answer_modified, tf_idf_context_modified,
        bm25_answer_modified, bm25_context_modified,
        vision_answer_modified, vision_context_modified,
        open_source_answer_modified, open_source_context_modified,
        tf_idf_bm25_open_RRF_Ranking_answer, tf_idf_bm25_open_RRF_Ranking_context,
        tf_idf_bm25_open_RRF_Ranking_modified_answer, tf_idf_bm25_open_RRF_Ranking_modified_context,
        tf_idf_bm25_open_RRF_Ranking_combined_answer, tf_idf_bm25_open_RRF_Ranking_combined_context,
        zeroShot, "Zero-shot doesn't have a context."
    )

# Gradio interface
def create_interface():
    with gr.Blocks() as interface:
        gr.Markdown("# Query Answering System")
        gr.Markdown("Enter a query to get the best model and the best answer using multiple retrieval models and ranking techniques.")
        query_input = gr.Textbox(label="Enter your query")
        
        with gr.Row():
            best_model_output = gr.Textbox(label="Best Model")
            best_answer_output = gr.Textbox(label="Best Answer")

        gr.Markdown("---")  # Horizontal line
        gr.Markdown("## All Answers and Contexts")

        def create_answer_row(label, context):
            with gr.Row():
                answer_textbox = gr.Textbox(label=f"{label} Answer", interactive=False).style(container=True)
                context_button = gr.Button(f"Show {label} Context")
                context_textbox = gr.Textbox(label=f"{label} Context", visible=False).style(container=True)

                context_button.click(
                    fn=lambda: gr.update(visible=True, value=context),
                    inputs=[],
                    outputs=context_textbox,
                )
                return answer_textbox, context_textbox
                

        with gr.Row():
            agent1_output, agent1_context_box = create_answer_row("Agent 1", agent1_context)
            agent2_output, agent2_context_box = create_answer_row("Agent 2", agent_text_context)
            boolean_output, boolean_context_box = create_answer_row("Boolean", boolean_context)
            tf_idf_output, tf_idf_context_box = create_answer_row("TF-IDF", tf_idf_context)
            bm25_output, bm25_context_box = create_answer_row("BM25", bm25_context)
            vision_output, vision_context_box = create_answer_row("Vision", vision_context)
            open_source_output, open_source_context_box = create_answer_row("Open Source", open_source_context)

        with gr.Row():
            boolean_mod_output, boolean_mod_context_box = create_answer_row("Boolean (Modified)", boolean_context_modified)
            tf_idf_mod_output, tf_idf_mod_context_box = create_answer_row("TF-IDF (Modified)", tf_idf_context_modified)
            bm25_mod_output, bm25_mod_context_box = create_answer_row("BM25 (Modified)", bm25_context_modified)
            vision_mod_output, vision_mod_context_box = create_answer_row("Vision (Modified)", vision_context_modified)
            open_source_mod_output, open_source_mod_context_box = create_answer_row("Open Source (Modified)", open_source_context_modified)

        with gr.Row():
            tf_idf_rrf_output, tf_idf_rrf_context_box = create_answer_row("TF-IDF + BM25 + Open RRF", tf_idf_bm25_open_RRF_Ranking_context)
            tf_idf_rrf_mod_output, tf_idf_rrf_mod_context_box = create_answer_row("TF-IDF + BM25 + Open RRF (Modified)", tf_idf_bm25_open_RRF_Ranking_modified_context)
            tf_idf_rrf_combined_output, tf_idf_rrf_combined_context_box = create_answer_row("TF-IDF + BM25 + Open RRF (Combined)", tf_idf_bm25_open_RRF_Ranking_combined_context)

        with gr.Row():
            zero_shot_output, zero_shot_context_box = create_answer_row("Zero Shot", "Zero-shot doesn't have a context.")

        gr.Button("Submit").click(
            fn=process_query,
            inputs=query_input,
            outputs=[
                best_model_output,
                best_answer_output,
                agent1_output, agent1_context_box,
                agent2_output, agent2_context_box,
                boolean_output, boolean_context_box,
                tf_idf_output, tf_idf_context_box,
                bm25_output, bm25_context_box,
                vision_output, vision_context_box,
                open_source_output, open_source_context_box,
                boolean_mod_output, boolean_mod_context_box,
                tf_idf_mod_output, tf_idf_mod_context_box,
                bm25_mod_output, bm25_mod_context_box,
                vision_mod_output, vision_mod_context_box,
                open_source_mod_output, open_source_context_box,
                tf_idf_rrf_output, tf_idf_rrf_context_box,
                tf_idf_rrf_mod_output, tf_idf_rrf_mod_context_box,
                tf_idf_rrf_combined_output, tf_idf_rrf_combined_context_box,
                zero_shot_output, zero_shot_context_box,
            ]
        )
    
    return interface

if __name__ == "__main__":
    create_interface().launch()