app.py

import gradio as gr
from pyserini.search.lucene import LuceneSearcher
import os
import json

def initialize_searcher(index_name):
    if not os.path.exists(index_name):
        os.system(f'python -c "from pyserini.search import LuceneSearcher; LuceneSearcher.from_prebuilt_index(\'{index_name}\')"')
    searcher = LuceneSearcher.from_prebuilt_index(index_name)
    searcher.set_bm25(k1=0.9, b=0.4)
    return searcher

def search_pyserini(query, top_k, index_name):
    try:
        searcher = initialize_searcher(index_name)
        hits = searcher.search(query, k=top_k)
        results = []
        for i, hit in enumerate(hits):
            doc = searcher.doc(hit.docid)
            doc_dict = json.loads(doc.raw())
            results.append({
                "rank": i + 1,
                "doc_id": hit.docid,
                "score": hit.score,
                "content": doc_dict['contents']
            })
        return format_results(results)
    except Exception as e:
        return f"<div class='error'>An error occurred: {str(e)}</div>"

def format_results(results):
    html = "<div class='results-container'>"
    for result in results:
        html += f"""
        <div class='result-item'>
            <h3>Rank {result['rank']} (Score: {result['score']:.4f})</h3>
            <p class='doc-id'>Doc ID: {result['doc_id']}</p>
            <p class='content'>{result['content']}</p>
        </div>
        """
    html += "</div>"
    return html

css = """
.gradio-container {
    font-family: 'Arial', sans-serif;
}
.results-container {
    display: flex;
    flex-direction: column;
    gap: 20px;
}
.result-item {
    border: 1px solid #ddd;
    border-radius: 8px;
    padding: 15px;
    width: 100%;
    box-shadow: 0 2px 4px rgba(0,0,0,0.1);
}
.result-item h3 {
    margin-top: 0;
    color: #333;
}
.doc-id {
    font-size: 0.9em;
    color: #666;
    margin-bottom: 10px;
}
.content {
    font-size: 0.95em;
    line-height: 1.4;
}
.error {
    color: red;
    font-weight: bold;
}
"""

with gr.Blocks(css=css) as iface:
    gr.Markdown("# Pyserini Search Interface")
    gr.Markdown("Enter a query to search using Pyserini with BM25 scoring (k1=0.9, b=0.4).")
    
    with gr.Row():
        index_input = gr.Textbox(
            value="msmarco-passage",
            lines=1,
            label="Prebuilt Index Name",
            placeholder="Enter the name of the prebuilt index"
        )
    
    with gr.Row():
        top_k_slider = gr.Slider(
            minimum=1,
            maximum=100,
            value=10,
            step=1,
            label="Number of top results to return"
        )
    
    with gr.Row():
        query_input = gr.Textbox(
            lines=1, 
            placeholder="Enter your search query here...",
            label="Search Query"
        )
    
    with gr.Row():
        search_button = gr.Button("Search", variant="primary")
    
    with gr.Row():
        output = gr.HTML(label="Search Results")
    
    search_button.click(
        fn=search_pyserini,
        inputs=[query_input, top_k_slider, index_input],
        outputs=output
    )

iface.launch()