app.py

import gradio as gr
import duckdb
from huggingface_hub import HfFileSystem
from huggingface_hub.hf_file_system import safe_quote
import pandas as pd

PARQUET_REVISION="refs/convert/parquet"
TABLE_WILDCARD="{table}"

fs = HfFileSystem()
duckdb.register_filesystem(fs)

def greet(dataset, config, split, sql):
    try:
        if TABLE_WILDCARD not in sql:
            raise Exception(f"Query must contains {TABLE_WILDCARD} wildcard.")
        # dataset="glue"
        # config="mnli"
        path=f"{config}/{dataset}-{split}.parquet" # Only from one split
        location=f"hf://datasets/{dataset}@{safe_quote(PARQUET_REVISION)}/{path}"
        print(location)
        sql = sql.replace(TABLE_WILDCARD, f"'{location}'")
        
        # result = duckdb.query(f"SELECT idx as id, premise as p FROM '{location}' LIMIT 2").to_df()
        result = duckdb.query(sql).to_df()
        print("QUERY SUCCESSED")
    except Exception as error:
        print(f"Error: {str(error)}")
        return pd.DataFrame({"Error": [f"❌ {str(error)}"]})
    return result

with gr.Blocks() as demo:
    gr.Markdown(" ## DuckDB demo using parquet revision")
    dataset = gr.Textbox(label="dataset", placeholder="mstz/iris")
    config = gr.Textbox(label="config", placeholder="iris")
    split = gr.Textbox(label="split", placeholder="train")
    sql = gr.Textbox(label="sql", placeholder=f"SELECT sepal_length FROM {TABLE_WILDCARD} LIMIT 3")
    run_button = gr.Button("Run")
    gr.Markdown("### Result")
    cached_responses_table = gr.DataFrame()
    run_button.click(greet, inputs=[dataset, config, split, sql], outputs=cached_responses_table)



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




# duckdb.query(f"SELECT idx as id, premise as p FROM '{location}' LIMIT 2").show()

# duckdb.query(f"SELECT idx as id, premise as p FROM '{location}' LIMIT 2")
# duckdb.query(f"SELECT max(idx) as max FROM '{location}' LIMIT 2")
# duckdb.query(f"SELECT idx  FROM '{location}'  ORDER BY idx DESC LIMIT 1").show()