app.py

import gradio as gr
import pandas as pd
import matplotlib.pyplot as plt
import io
import google.generativeai as genai
from PIL import Image
import ast
import re

def process_file(api_key, file, instructions):
    genai.configure(api_key=api_key)
    model = genai.GenerativeModel('gemini-2.5-pro-preview-03-25')

    try:
        df = pd.read_csv(file.name) if file.name.endswith('.csv') else pd.read_excel(file.name)
        if df.empty:
            raise ValueError("Uploaded file contains no data")
    except Exception as e:
        print(f"Data Error: {str(e)}")
        return [generate_error_image(str(e))]*3

    # Enhanced prompt with strict plotting requirements
    prompt = f"""Generate 3 matplotlib codes with these rules:
    1. Use ONLY these variables: df (DataFrame), plt
    2. Each visualization MUST:
       - Plot actual data from df
       - Include title, axis labels, and data labels if needed
       - Use clear color schemes
       - Avoid empty plots
    3. Code structure:
       plt.figure(figsize=(16,9), dpi=120)
       plt.style.use('ggplot')
       # Plotting code using df columns: {list(df.columns)}
       plt.tight_layout()
    
    Sample data: {df.head(3).to_dict()}
    User instructions: {instructions or 'General insights'}
    
    Format EXACTLY as:
    # Visualization 1
    [complete code]
    """

    try:
        response = model.generate_content(prompt)
        code_blocks = re.split(r'# Visualization \d+', response.text)[1:4]
    except Exception as e:
        return [generate_error_image("API Error")]*3

    visualizations = []
    for i, block in enumerate(code_blocks, 1):
        try:
            # Advanced code sanitization
            cleaned_code = sanitize_code(block, df.columns)
            
            # Validate and execute
            ast.parse(cleaned_code)
            img = execute_plot_code(cleaned_code, df)
            visualizations.append(img)
        except Exception as e:
            print(f"Visualization {i} Error: {str(e)}")
            visualizations.append(generate_error_image(f"Plot {i} Error"))

    return visualizations + [generate_error_image("Not Generated")]*(3-len(visualizations))

def sanitize_code(code_block, columns):
    """Clean and validate generated code"""
    replacements = {
        r"'y_axis'": f"'{columns[1]}'" if len(columns) > 1 else "'Value'",
        r"'x_axis'": f"'{columns[0]}'",
        r"data": "df",
        r"plt.legend\(\)": ""  # Remove empty legend calls
    }
    
    cleaned = []
    for line in code_block.split('\n'):
        line = line.strip()
        if not line or line.startswith('`'):
            continue
            
        # Apply replacements
        for pattern, replacement in replacements.items():
            line = re.sub(pattern, replacement, line)
            
        cleaned.append(line)
    
    return '\n'.join(cleaned)

def execute_plot_code(code, df):
    """Safely execute plotting code"""
    buf = io.BytesIO()
    plt.figure(figsize=(16, 9), dpi=120)
    plt.style.use('ggplot')
    
    try:
        exec(code, {'df': df, 'plt': plt})
        plt.tight_layout()
        plt.savefig(buf, format='png', bbox_inches='tight')
        buf.seek(0)
        return Image.open(buf)
    finally:
        plt.close()

def generate_error_image(message):
    """Create error indication image"""
    img = Image.new('RGB', (1920, 1080), color=(73, 109, 137))
    return img

# Gradio interface
with gr.Blocks(theme=gr.themes.Default(spacing_size="lg")) as demo:
    gr.Markdown("# Professional Data Visualizer")
    
    with gr.Row():
        api_key = gr.Textbox(label="Gemini API Key", type="password")
        file = gr.File(label="Upload Data File", file_types=[".csv", ".xlsx"])
    
    instructions = gr.Textbox(label="Visualization Instructions")
    submit = gr.Button("Generate Insights", variant="primary")
    
    with gr.Row():
        outputs = [gr.Image(label=f"Visualization {i+1}", width=600) for i in range(3)]

    submit.click(
        process_file,
        inputs=[api_key, file, instructions],
        outputs=outputs
    )

demo.launch()