import gradio as gr import pandas as pd import numpy as np import plotly.express as px import io import json import warnings import google.generativeai as genai import os from typing import List, Dict, Any, Tuple, Optional import re # --- Configuration & Constants --- warnings.filterwarnings('ignore') MAX_DASHBOARD_PLOTS = 10 CSS = """ #app-title { text-align: center; font-weight: 800; font-size: 2.5rem; color: #f9fafb; padding-top: 10px; } .stat-card { border-radius: 12px !important; padding: 20px !important; background: #1f2937 !important; border: 1px solid #374151 !important; text-align: center; transition: all 0.3s ease; } .stat-card:hover { transform: translateY(-5px); box-shadow: 0 10px 15px -3px rgba(0,0,0,0.1), 0 4px 6px -2px rgba(0,0,0,0.05); } .stat-card-title { font-size: 16px; font-weight: 500; color: #9ca3af !important; margin-bottom: 8px; } .stat-card-value { font-size: 32px; font-weight: 700; color: #f9fafb !important; } .sidebar { background-color: #111827 !important; padding: 15px; border-right: 1px solid #374151 !important; min-height: 100vh; } .sidebar .gr-button { width: 100%; text-align: left !important; background: none !important; border: none !important; box-shadow: none !important; color: #d1d5db !important; font-size: 16px !important; padding: 12px 10px !important; margin-bottom: 8px !important; border-radius: 8px !important; transition: background-color 0.2s ease; } .sidebar .gr-button:hover { background-color: #374151 !important; } .sidebar .gr-button.selected { background-color: #4f46e5 !important; font-weight: 600 !important; color: white !important; } .explanation-block { background-color: #1e3a8a !important; border-left: 4px solid #3b82f6 !important; padding: 12px; color: #e5e7eb !important; border-radius: 4px; } """ class DataExplorerApp: """A professional-grade, AI-powered data exploration application.""" def __init__(self): self.demo = self._build_ui() def _build_ui(self) -> gr.Blocks: with gr.Blocks(theme=gr.themes.Glass(primary_hue="indigo", secondary_hue="blue"), css=CSS, title="AI Data Explorer Pro") as demo: state_var = gr.State({}) # Component Definition cockpit_btn = gr.Button("📊 Data Cockpit", elem_classes="selected", elem_id="cockpit") deep_dive_btn = gr.Button("🔍 Deep Dive Builder", elem_id="deep_dive") copilot_btn = gr.Button("🤖 Chief Data Scientist", elem_id="co-pilot") file_input = gr.File(label="📁 Upload CSV File", file_types=[".csv"]) status_output = gr.Markdown("Status: Awaiting data...") api_key_input = gr.Textbox(label="🔑 Gemini API Key", type="password", placeholder="Enter key to enable AI...") suggestion_btn = gr.Button("Get Smart Suggestions", variant="secondary", interactive=False) rows_stat, cols_stat = gr.Textbox("0", interactive=False, show_label=False), gr.Textbox("0", interactive=False, show_label=False) quality_stat, time_cols_stat = gr.Textbox("0%", interactive=False, show_label=False), gr.Textbox("0", interactive=False, show_label=False) suggestion_buttons = [gr.Button(visible=False) for _ in range(5)] plot_type_dd = gr.Dropdown(['histogram', 'bar', 'scatter', 'box'], label="Plot Type", value='histogram') x_col_dd = gr.Dropdown([], label="X-Axis / Column", interactive=False) y_col_dd = gr.Dropdown([], label="Y-Axis (for Scatter/Box)", visible=False, interactive=False) add_plot_btn, clear_plots_btn = gr.Button("Add to Dashboard", variant="primary", interactive=False), gr.Button("Clear Dashboard", interactive=False) dashboard_plots = [gr.Plot(visible=False) for _ in range(MAX_DASHBOARD_PLOTS)] chatbot = gr.Chatbot(height=500, label="Conversation", show_copy_button=True, avatar_images=(None, "bot.png")) copilot_explanation, copilot_code = gr.Markdown(visible=False, elem_classes="explanation-block"), gr.Code(language="python", visible=False, label="Executed Code") copilot_plot, copilot_table = gr.Plot(visible=False, label="Generated Visualization"), gr.Dataframe(visible=False, label="Generated Table", wrap=True) chat_input, chat_submit_btn = gr.Textbox(label="Your Question", placeholder="e.g., 'What is the relationship between age and salary?'", scale=4), gr.Button("Ask AI", variant="primary", interactive=False) # Layout Arrangement with gr.Row(): with gr.Column(scale=1, elem_classes="sidebar"): gr.Markdown("## 🚀 AI Explorer Pro", elem_id="app-title"); cockpit_btn; deep_dive_btn; copilot_btn; gr.Markdown("---") file_input; status_output; gr.Markdown("---"); api_key_input; suggestion_btn with gr.Column(scale=4): welcome_page, cockpit_page, deep_dive_page, copilot_page = [gr.Column(visible=i==0) for i in range(4)] with welcome_page: gr.Markdown("# Welcome to the AI Data Explorer Pro\n> Please **upload a CSV file** and **enter your Gemini API key** to begin your analysis.") with cockpit_page: gr.Markdown("## 📊 Data Cockpit: At-a-Glance Overview") with gr.Row(): for title, stat_comp in [("Rows", rows_stat), ("Columns", cols_stat), ("Data Quality", quality_stat), ("Date/Time Cols", time_cols_stat)]: with gr.Column(elem_classes="stat-card"): gr.Markdown(f"
{title}
"); stat_comp with gr.Accordion(label="✨ AI Smart Suggestions", open=True): [btn for btn in suggestion_buttons] with deep_dive_page: gr.Markdown("## 🔍 Deep Dive: Manual Dashboard Builder"); gr.Markdown("Construct visualizations to investigate specific relationships.") with gr.Row(): plot_type_dd; x_col_dd; y_col_dd with gr.Row(): add_plot_btn; clear_plots_btn with gr.Column(): [plot for plot in dashboard_plots] with copilot_page: gr.Markdown("## 🤖 Chief Data Scientist: Your AI Partner"); chatbot with gr.Accordion("AI's Detailed Response", open=True): copilot_explanation; copilot_code; copilot_plot; copilot_table with gr.Row(): chat_input; chat_submit_btn # Event Handlers Registration pages, nav_buttons = [welcome_page, cockpit_page, deep_dive_page, copilot_page], [cockpit_btn, deep_dive_btn, copilot_btn] for i, btn in enumerate(nav_buttons): btn.click(lambda id=btn.elem_id: self._switch_page(id, pages), outputs=pages).then( lambda i=i: [gr.update(elem_classes="selected" if j==i else "") for j in range(len(nav_buttons))], outputs=nav_buttons) file_input.upload(self.load_and_process_file, inputs=[file_input], outputs=[ state_var, status_output, *pages, rows_stat, cols_stat, quality_stat, time_cols_stat, x_col_dd, y_col_dd, add_plot_btn]) api_key_input.change(lambda x: gr.update(interactive=bool(x)), inputs=[api_key_input], outputs=[suggestion_btn]) chat_input.change(lambda x: gr.update(interactive=bool(x.strip())), inputs=[chat_input], outputs=[chat_submit_btn]) plot_type_dd.change(self._update_plot_controls, inputs=[plot_type_dd], outputs=[y_col_dd]) add_plot_btn.click(self.add_plot_to_dashboard, inputs=[state_var, x_col_dd, y_col_dd, plot_type_dd], outputs=[state_var, clear_plots_btn, *dashboard_plots]) clear_plots_btn.click(self.clear_dashboard, inputs=[state_var], outputs=[state_var, clear_plots_btn, *dashboard_plots]) suggestion_btn.click(self.get_ai_suggestions, inputs=[state_var, api_key_input], outputs=suggestion_buttons) for btn in suggestion_buttons: btn.click(self.handle_suggestion_click, inputs=[btn], outputs=[*pages, chat_input]) chat_submit_btn.click(self.respond_to_chat, [state_var, api_key_input, chat_input, chatbot], [chatbot, copilot_explanation, copilot_code, copilot_plot, copilot_table]).then(lambda: "", outputs=[chat_input]) chat_input.submit(self.respond_to_chat, [state_var, api_key_input, chat_input, chatbot], [chatbot, copilot_explanation, copilot_code, copilot_plot, copilot_table]).then(lambda: "", outputs=[chat_input]) return demo def launch(self): self.demo.launch(debug=True) # --- Backend Logic Methods --- def _switch_page(self, page_id: str, all_pages: List) -> List[gr.update]: visibility = {"welcome":0, "cockpit":1, "deep_dive":2, "co-pilot":3} return [gr.update(visible=i == visibility.get(page_id, 0)) for i in range(len(all_pages))] def _update_plot_controls(self, plot_type: str) -> gr.update: return gr.update(visible=plot_type in ['scatter', 'box']) def load_and_process_file(self, file_obj: Any) -> Tuple[Any, ...]: try: df = pd.read_csv(file_obj.name, low_memory=False) metadata = self._extract_dataset_metadata(df) state = {'df': df, 'metadata': metadata, 'dashboard_plots': []} rows, cols, quality = metadata['shape'][0], metadata['shape'][1], metadata['data_quality'] page_updates = self._switch_page("cockpit", [0,1,2,3]) return (state, f"✅ **{os.path.basename(file_obj.name)}** loaded.", *page_updates, f"{rows:,}", f"{cols}", f"{quality}%", f"{len(metadata['datetime_cols'])}", gr.update(choices=metadata['columns'], interactive=True), gr.update(choices=metadata['columns'], interactive=True), gr.update(interactive=True)) except Exception as e: gr.Error(f"File Load Error: {e}"); page_updates = self._switch_page("welcome", [0,1,2,3]); return {}, f"❌ Error: {e}", *page_updates, "0", "0", "0%", "0", gr.update(choices=[], interactive=False), gr.update(choices=[], interactive=False), gr.update(interactive=False) def _extract_dataset_metadata(self, df: pd.DataFrame) -> Dict[str, Any]: rows, cols, quality = df.shape[0], df.shape[1], round((df.notna().sum().sum() / (df.size)) * 100, 1) if df.size > 0 else 0 return {'shape': (rows, cols), 'columns': df.columns.tolist(), 'numeric_cols': df.select_dtypes(include=np.number).columns.tolist(), 'categorical_cols': df.select_dtypes(include=['object', 'category']).columns.tolist(), 'datetime_cols': df.select_dtypes(include=['datetime64', 'datetime64[ns]']).columns.tolist(), 'dtypes_head': df.head(3).to_string(), 'data_quality': quality} def add_plot_to_dashboard(self, state: Dict, x_col: str, y_col: Optional[str], plot_type: str) -> List[Any]: dashboard_plots = state.get('dashboard_plots', []) if len(dashboard_plots) >= MAX_DASHBOARD_PLOTS: gr.Warning(f"Dashboard is full. Max {MAX_DASHBOARD_PLOTS} plots."); return [state, gr.update(interactive=True), *self._get_plot_updates(state)] if not x_col: gr.Warning("Please select an X-axis column."); return [state, gr.update(interactive=True), *self._get_plot_updates(state)] df, title = state.get('df'), f"{plot_type.capitalize()}: {y_col} by {x_col}" if y_col and plot_type in ['box', 'scatter'] else f"Distribution of {x_col}" try: fig=None; if plot_type == 'histogram': fig = px.histogram(df, x=x_col, title=title) elif plot_type == 'box': fig = px.box(df, x=x_col, y=y_col, title=title) elif plot_type == 'scatter': fig = px.scatter(df, x=x_col, y=y_col, title=title, trendline="ols") elif plot_type == 'bar': fig = px.bar(df[x_col].value_counts().nlargest(20), title=f"Top 20 for {x_col}") if fig: fig.update_layout(template="plotly_dark"); dashboard_plots.append(fig); gr.Info(f"Added '{title}' to dashboard.") return [state, gr.update(interactive=True), *self._get_plot_updates(state)] except Exception as e: gr.Error(f"Plotting Error: {e}"); return [state, gr.update(interactive=True), *self._get_plot_updates(state)] def _get_plot_updates(self, state: Dict) -> List[gr.update]: plots = state.get('dashboard_plots', []) return [gr.update(value=plots[i] if i < len(plots) else None, visible=i < len(plots)) for i in range(MAX_DASHBOARD_PLOTS)] def clear_dashboard(self, state: Dict) -> List[Any]: state['dashboard_plots'] = []; gr.Info("Dashboard cleared."); return [state, gr.update(interactive=False), *self._get_plot_updates(state)] def get_ai_suggestions(self, state: Dict, api_key: str) -> List[gr.update]: if not api_key: gr.Warning("API Key is required."); return [gr.update(visible=False)]*5 if not state: gr.Warning("Please load data first."); return [gr.update(visible=False)]*5 metadata, columns = state.get('metadata', {}), state.get('metadata', {}).get('columns', []) prompt = f"From columns {columns}, generate 4 impactful analytical questions. Return ONLY a JSON list of strings." try: genai.configure(api_key=api_key); suggestions = json.loads(genai.GenerativeModel('gemini-1.5-flash').generate_content(prompt).text) return [gr.Button(s, visible=True) for s in suggestions] + [gr.Button(visible=False)] * (5 - len(suggestions)) except Exception as e: gr.Error(f"AI Suggestion Error: {e}"); return [gr.update(visible=False)]*5 def handle_suggestion_click(self, question: str) -> Tuple[gr.update, ...]: return *self._switch_page("co-pilot", [0,1,2,3]), question def _sanitize_and_parse_json(self, raw_text: str) -> Dict: """Cleans and parses a JSON string from an LLM response.""" # Remove markdown code blocks clean_text = re.sub(r'```json\n?|```', '', raw_text).strip() # Escape single backslashes that are not already escaped clean_text = re.sub(r'(? Any: if not user_message.strip(): return history, *[gr.update()]*4 if not api_key or not state: history.append((user_message, "I need a Gemini API key and a dataset to work.")); return history, *[gr.update(visible=False)]*4 history.append((user_message, "Thinking... 🤔")); yield history, *[gr.update(visible=False)]*4 metadata, dtypes_head = state.get('metadata', {}), state.get('metadata', {}).get('dtypes_head', 'No metadata available.') prompt = f"""You are 'Chief Data Scientist', an expert AI analyst. Your goal is to answer a user's question about a pandas DataFrame (`df`) by writing and executing Python code. **Instructions:** 1. **Analyze:** Understand the user's intent. Infer the best plot type. 2. **Plan:** Briefly explain your plan. 3. **Code:** Write Python code. Use `fig` for plots (`template='plotly_dark'`) and `result_df` for tables. 4. **Insight:** Provide a one-sentence business insight. 5. **Respond ONLY with a single JSON object with keys: "plan", "code", "insight".** **Metadata:** {dtypes_head} **User Question:** "{user_message}" """ try: genai.configure(api_key=api_key) # CRITICAL FIX: Use the new sanitizer function response_json = self._sanitize_and_parse_json(genai.GenerativeModel('gemini-1.5-flash').generate_content(prompt).text) plan, code, insight = response_json.get("plan"), response_json.get("code"), response_json.get("insight") stdout, fig, df_result, error = self._safe_exec(code, {'df': state['df'], 'px': px, 'pd': pd}) history[-1] = (user_message, f"**Plan:** {plan}") explanation = f"**Insight:** {insight}" if stdout: explanation += f"\n\n**Console Output:**\n```\n{stdout}\n```" if error: gr.Error(f"AI Code Execution Failed: {error}") yield (history, gr.update(visible=bool(explanation), value=explanation), gr.update(visible=bool(code), value=code), gr.update(visible=bool(fig), value=fig), gr.update(visible=bool(df_result is not None), value=df_result)) except Exception as e: history[-1] = (user_message, f"I encountered an error processing the AI response. Please rephrase your question.\n\n**Details:** `{str(e)}`") yield history, *[gr.update(visible=False)]*4 def _safe_exec(self, code_string: str, local_vars: Dict) -> Tuple[Any, ...]: try: output_buffer = io.StringIO() with redirect_stdout(output_buffer): exec(code_string, globals(), local_vars) return output_buffer.getvalue(), local_vars.get('fig'), local_vars.get('result_df'), None except Exception as e: return None, None, None, str(e) if __name__ == "__main__": if not os.path.exists("bot.png"): try: from PIL import Image Image.new('RGB', (1, 1)).save('bot.png') except ImportError: print("Pillow not installed, cannot create dummy bot.png.") app = DataExplorerApp() app.launch()