app.py

import json
import gradio as gr
import duckdb
from functools import lru_cache
import pandas as pd
import plotly.express as px
import openai
import os

# =========================
# Configuration and Setup
# =========================

# Load the Parquet dataset path
dataset_path = 'sample_contract_df.parquet'  # Update with your Parquet file path

# Provided schema
schema = [
    {"column_name": "department_ind_agency", "column_type": "VARCHAR"},
    {"column_name": "cgac", "column_type": "BIGINT"},
    {"column_name": "sub_tier", "column_type": "VARCHAR"},
    {"column_name": "fpds_code", "column_type": "VARCHAR"},
    {"column_name": "office", "column_type": "VARCHAR"},
    {"column_name": "aac_code", "column_type": "VARCHAR"},
    {"column_name": "posteddate", "column_type": "VARCHAR"},
    {"column_name": "type", "column_type": "VARCHAR"},
    {"column_name": "basetype", "column_type": "VARCHAR"},
    {"column_name": "popstreetaddress", "column_type": "VARCHAR"},
    {"column_name": "popcity", "column_type": "VARCHAR"},
    {"column_name": "popstate", "column_type": "VARCHAR"},
    {"column_name": "popzip", "column_type": "VARCHAR"},
    {"column_name": "popcountry", "column_type": "VARCHAR"},
    {"column_name": "active", "column_type": "VARCHAR"},
    {"column_name": "awardnumber", "column_type": "VARCHAR"},
    {"column_name": "awarddate", "column_type": "VARCHAR"},
    {"column_name": "award", "column_type": "DOUBLE"},
    {"column_name": "awardee", "column_type": "VARCHAR"},
    {"column_name": "state", "column_type": "VARCHAR"},
    {"column_name": "city", "column_type": "VARCHAR"},
    {"column_name": "zipcode", "column_type": "VARCHAR"},
    {"column_name": "countrycode", "column_type": "VARCHAR"}
]

# Cache the schema loading
@lru_cache(maxsize=1)
def get_schema():
    return schema

# Map column names to their types
COLUMN_TYPES = {col['column_name']: col['column_type'] for col in get_schema()}

# =========================
# Database Interaction
# =========================

def load_dataset_schema():
    """
    Loads the dataset schema into DuckDB by creating a view.
    """
    con = duckdb.connect()
    try:
        # Drop the view if it exists to avoid errors
        con.execute("DROP VIEW IF EXISTS contract_data")
        con.execute(f"CREATE VIEW contract_data AS SELECT * FROM '{dataset_path}'")
        return True
    except Exception as e:
        print(f"Error loading dataset schema: {e}")
        return False
    finally:
        con.close()

# =========================
# OpenAI API Integration
# =========================

async def parse_query(nl_query):
    """
    Converts a natural language query into a SQL query using OpenAI's GPT-4-turbo model.
    """
    messages = [
        {"role": "system", "content": (
            "You are an assistant that converts natural language queries into SQL queries "
            "for a DuckDB database named 'contract_data'. Use the provided schema to form accurate SQL queries."
        )},
        {"role": "user", "content": (
            f"Schema:\n{json.dumps(schema, indent=2)}\n\nNatural Language Query:\n\"{nl_query}\"\n\nSQL Query:"
        )}
    ]

    try:
        response = await openai.ChatCompletion.acreate(
            model="gpt-3.5-turbo",
            messages=messages,
            temperature=0,
            max_tokens=150,
        )
        sql_query = response.choices[0].message['content'].strip()
        return sql_query
    except Exception as e:
        return f"Error generating SQL query: {e}"

# =========================
# Plotting Utilities
# =========================

def detect_plot_intent(nl_query):
    """
    Detects if the user's query involves plotting based on the presence of specific keywords.
    """
    plot_keywords = [
        'plot', 'graph', 'chart', 'distribution', 'visualize', 'histogram',
        'bar chart', 'line chart', 'scatter plot', 'pie chart'
    ]
    for keyword in plot_keywords:
        if keyword in nl_query.lower():
            return True
    return False

async def generate_sql_and_plot_code(query):
    """
    Generates SQL query and plotting code based on the natural language input.
    """
    is_plot = detect_plot_intent(query)
    sql_query = await parse_query(query)
    plot_code = ""
    if is_plot and not sql_query.startswith("Error"):
        # Generate plot code based on the query
        plot_code = """
import plotly.express as px
fig = px.bar(result_df, x='x_column', y='y_column', title='Generated Plot')
fig.update_layout(title_x=0.5)
"""
    return sql_query, plot_code

def execute_query(sql_query):
    """
    Executes the SQL query and returns results or an error message.
    """
    if sql_query.startswith("Error"):
        return None, sql_query  # Pass the error message forward

    try:
        con = duckdb.connect()
        con.execute(f"CREATE OR REPLACE VIEW contract_data AS SELECT * FROM '{dataset_path}'")
        result_df = con.execute(sql_query).fetchdf()
        con.close()
        return result_df, ""
    except Exception as e:
        return None, f"Error executing query: {e}"

def generate_plot(plot_code, result_df):
    """
    Executes the plot code to generate a plot from the result DataFrame.
    """
    if not plot_code.strip():
        return None, "No plot code provided."
    try:
        if result_df.empty:
            return None, "Result DataFrame is empty."
        columns = result_df.columns.tolist()
        if len(columns) < 2:
            return None, "Not enough columns to plot."
        plot_code = plot_code.replace('x_column', columns[0])
        plot_code = plot_code.replace('y_column', columns[1])

        local_vars = {'result_df': result_df, 'px': px}
        exec(plot_code, {}, local_vars)
        fig = local_vars.get('fig', None)
        return fig, "" if fig else "Plot could not be generated."
    except Exception as e:
        return None, f"Error generating plot: {e}"

# =========================
# Schema Display
# =========================

@lru_cache(maxsize=1)
def get_schema_json():
    return json.dumps(get_schema(), indent=2)

# =========================
# Initialize Dataset Schema
# =========================

if not load_dataset_schema():
    raise Exception("Failed to load dataset schema. Please check the dataset path and format.")

# =========================
# Gradio Application UI
# =========================

with gr.Blocks() as demo:
    gr.Markdown("""
    # Parquet SQL Query and Plotting App

    **Query and visualize data** in `sample_contract_df.parquet`
    """)

    with gr.Tabs():
        with gr.TabItem("Query Data"):
            with gr.Row():
                with gr.Column(scale=1):
                    query = gr.Textbox(
                        label="Natural Language Query",
                        placeholder='e.g., "Show all awards greater than 1,000,000 in California"',
                        lines=4
                    )
                    btn_generate = gr.Button("Generate SQL")
                    sql_out = gr.Code(label="Generated SQL Query", language="sql")
                    plot_code_out = gr.Code(label="Generated Plot Code", language="python")
                    btn_execute = gr.Button("Execute Query")
                    error_out = gr.Markdown("", visible=False)
                with gr.Column(scale=2):
                    results_out = gr.Dataframe(label="Query Results", interactive=False)
                    plot_out = gr.Plot(label="Plot")

        with gr.TabItem("Dataset Schema"):
            gr.Markdown("### Dataset Schema")
            schema_display = gr.JSON(label="Schema", value=json.loads(get_schema_json()))

    async def on_generate_click(nl_query):
        sql_query, plot_code = await generate_sql_and_plot_code(nl_query)
        return sql_query, plot_code

    def on_execute_click(sql_query, plot_code):
        result_df, error_msg = execute_query(sql_query)
        if error_msg:
            return None, None, error_msg
        if plot_code.strip():
            fig, plot_error = generate_plot(plot_code, result_df)
            return result_df, fig, plot_error if plot_error else ""
        else:
            return result_df, None, ""

    btn_generate.click(
        fn=on_generate_click,
        inputs=query,
        outputs=[sql_out, plot_code_out],
    )
    btn_execute.click(
        fn=on_execute_click,
        inputs=[sql_out, plot_code_out],
        outputs=[results_out, plot_out, error_out],
    )

demo.launch()