app.py

from transformers import Tool, ReactCodeAgent, HfApiEngine
import gradio as gr
import pandas as pd
import numpy as np
import plotly.express as px
import plotly.graph_objects as go
from typing import Dict, List, Optional
import openai
import seaborn as sns
import matplotlib.pyplot as plt
import io
import base64

# Custom Tools for Data Analysis
class DataVisualizationTool(Tool):
    name = "data_visualizer"
    description = """Creates various types of visualizations from data:
    - Correlation heatmaps
    - Distribution plots
    - Scatter plots
    - Time series plots
    Returns the plots as base64 encoded images."""

    inputs = {
        "data": {
            "type": "dict",
            "description": "DataFrame as dictionary"
        },
        "plot_type": {
            "type": "string",
            "description": "Type of plot to create: 'heatmap', 'distribution', 'scatter'"
        },
        "columns": {
            "type": "list",
            "description": "List of columns to plot"
        }
    }
    output_type = "string"  # base64 encoded image

    def forward(self, data: Dict, plot_type: str, columns: List[str]) -> str:
        df = pd.DataFrame(data)
        plt.figure(figsize=(10, 6))
        
        if plot_type == "heatmap":
            sns.heatmap(df[columns].corr(), annot=True, cmap='coolwarm')
            plt.title("Correlation Heatmap")
        elif plot_type == "distribution":
            for col in columns:
                sns.histplot(df[col], kde=True, label=col)
            plt.title("Distribution Plot")
            plt.legend()
        elif plot_type == "scatter":
            if len(columns) >= 2:
                sns.scatterplot(data=df, x=columns[0], y=columns[1])
                plt.title(f"Scatter Plot: {columns[0]} vs {columns[1]}")

        # Convert plot to base64
        buf = io.BytesIO()
        plt.savefig(buf, format='png')
        plt.close()
        buf.seek(0)
        return base64.b64encode(buf.read()).decode('utf-8')

class DataAnalysisTool(Tool):
    name = "data_analyzer"
    description = """Performs statistical analysis on data:
    - Basic statistics (mean, median, std)
    - Correlation analysis
    - Missing value analysis
    - Outlier detection"""

    inputs = {
        "data": {
            "type": "dict",
            "description": "DataFrame as dictionary"
        },
        "analysis_type": {
            "type": "string",
            "description": "Type of analysis: 'basic', 'correlation', 'missing', 'outliers'"
        },
        "columns": {
            "type": "list",
            "description": "List of columns to analyze"
        }
    }
    output_type = "dict"

    def forward(self, data: Dict, analysis_type: str, columns: List[str]) -> Dict:
        df = pd.DataFrame(data)
        selected_cols = [col for col in columns if col in df.columns]
        
        if analysis_type == "basic":
            return {
                "statistics": df[selected_cols].describe().to_dict(),
                "skew": df[selected_cols].skew().to_dict(),
                "kurtosis": df[selected_cols].kurtosis().to_dict()
            }
        elif analysis_type == "correlation":
            numeric_cols = df[selected_cols].select_dtypes(include=[np.number])
            return {
                "correlation": numeric_cols.corr().to_dict(),
                "covariance": numeric_cols.cov().to_dict()
            }
        elif analysis_type == "missing":
            return {
                "missing_counts": df[selected_cols].isnull().sum().to_dict(),
                "missing_percentages": (df[selected_cols].isnull().mean() * 100).to_dict()
            }
        elif analysis_type == "outliers":
            outliers = {}
            for col in selected_cols:
                if df[col].dtype in [np.float64, np.int64]:
                    Q1 = df[col].quantile(0.25)
                    Q3 = df[col].quantile(0.75)
                    IQR = Q3 - Q1
                    outliers[col] = {
                        "outliers_count": len(df[(df[col] < Q1 - 1.5 * IQR) | (df[col] > Q3 + 1.5 * IQR)]),
                        "lower_bound": Q1 - 1.5 * IQR,
                        "upper_bound": Q3 + 1.5 * IQR
                    }
            return {"outliers": outliers}

def create_demo():
    # Initialize tools
    viz_tool = DataVisualizationTool()
    analysis_tool = DataAnalysisTool()
    
    # Create agent with tools
    llm_engine = HfApiEngine()  # Uses default model
    agent = ReactCodeAgent(
        tools=[viz_tool, analysis_tool],
        llm_engine=llm_engine
    )

    with gr.Blocks(theme=gr.themes.Soft()) as demo:
        gr.Markdown("# 🔬 Advanced Data Analysis Agent")
        
        with gr.Row():
            with gr.Column():
                api_key = gr.Textbox(
                    label="OpenAI API Key",
                    type="password",
                    placeholder="sk-..."
                )
                file_input = gr.File(
                    label="Upload CSV",
                    file_types=[".csv"]
                )
                with gr.Accordion("Advanced Settings", open=False):
                    system_prompt = gr.Textbox(
                        label="System Prompt",
                        value="""You are a data science expert. Analyze the data and create 
                               visualizations to help understand patterns and insights.""",
                        lines=3
                    )
            
            with gr.Column():
                chat = gr.Chatbot(label="Analysis Chat")
                msg = gr.Textbox(
                    label="Ask about your data",
                    placeholder="What insights can you find in this dataset?"
                )
                clear = gr.Button("Clear")
        
        # State for storing the DataFrame
        df_state = gr.State(None)
        
        def process_file(file):
            if file is None:
                return None
            return pd.read_csv(file.name)
        
        def process_message(message, chat_history, api_key, df):
            if df is None:
                return chat_history + [(message, "Please upload a CSV file first.")]
                
            try:
                # Convert DataFrame to dict for tools
                data_dict = df.to_dict()
                
                # Get all columns for potential analysis
                columns = list(df.columns)
                
                # Use agent to analyze and create visualizations
                response = agent.run(
                    f"""Analyze this data: {message}
                    Available columns: {columns}
                    Use the data_analyzer and data_visualizer tools to create insights."""
                )
                
                return chat_history + [(message, response)]
                
            except Exception as e:
                return chat_history + [(message, f"Error: {str(e)}")]
        
        file_input.change(
            process_file,
            inputs=[file_input],
            outputs=[df_state]
        )
        
        msg.submit(
            process_message,
            inputs=[msg, chat, api_key, df_state],
            outputs=[chat]
        )
        
        clear.click(lambda: None, None, chat)
        
        return demo

if __name__ == "__main__":
    demo = create_demo()
    demo.launch()
else:
    demo.launch(show_api=False)