app.py

import os
import sys
import subprocess
import requests
import gradio as gr
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np

from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.preprocessing import LabelEncoder

# --------------------------------------------------------------------------------
# OPTIONAL: dynamic installation for rarely used packages not in requirements.txt
# --------------------------------------------------------------------------------

def install_library(library):
    """
    Install a library using pip.
    Useful for rarely used packages NOT in requirements.txt.
    """
    try:
        subprocess.check_call([sys.executable, "-m", "pip", "install", library])
        return f"Successfully installed {library}."
    except Exception as e:
        return f"Error installing {library}: {str(e)}"

def dynamic_import(library, alias=None):
    """
    Dynamically import a library. If not found, try to install it, then import again.
    """
    try:
        if alias:
            globals()[alias] = __import__(library)
        else:
            globals()[library] = __import__(library)
    except ImportError:
        install_msg = install_library(library)
        print(install_msg)
        globals()[library] = __import__(library)


# --------------------------------------------------------------------------------
# LLM CALLS: GPT-4o-mini, OpenAI, DeepSeek, Gemini
# --------------------------------------------------------------------------------
import openai
from huggingface_hub import InferenceClient

def call_gpt4o_mini(api_key, user_prompt):
    """
    Calls a GPT-4o-mini model hosted on Hugging Face.
    Replace 'someUser/gpt-4o-mini' with your actual model repo.
    """
    if not api_key:
        return "No Hugging Face API key provided. Cannot call GPT-4o-mini."

    try:
        client = InferenceClient(
            repo_id="someUser/gpt-4o-mini",  # <--- Replace with your real GPT-4o-mini repo
            token=api_key
        )
        # We use text_generation endpoint; adapt if your model differs
        response = client.text_generation(user_prompt, max_new_tokens=128)
        # 'response' can be a string or dict depending on the endpoint. Assume it's a string:
        return response
    except Exception as e:
        return f"Error calling GPT-4o-mini: {str(e)}"

def call_openai(api_key, user_prompt):
    """Calls OpenAI's API (example usage)."""
    openai.api_key = api_key
    try:
        response = openai.Completion.create(
            model="text-davinci-003",
            prompt=user_prompt,
            max_tokens=128
        )
        return response["choices"][0]["text"].strip()
    except Exception as e:
        return f"OpenAI Error: {str(e)}"

def call_deepseek(api_key, user_prompt):
    """
    Hypothetical function to call a DeepSeek API endpoint.
    Replace with real DeepSeek logic as needed.
    """
    try:
        headers = {
            "Content-Type": "application/json",
            "Authorization": f"Bearer {api_key}"
        }
        payload = {
            "prompt": user_prompt,
            "max_tokens": 128
        }
        # Example POST; adapt to the real DeepSeek endpoint
        response = requests.post(
            "https://api.deepseek.ai/v1/chat",
            json=payload,
            headers=headers
        )
        response.raise_for_status()
        data = response.json()
        return data["choices"][0]["text"].strip()
    except Exception as e:
        return f"DeepSeek Error: {str(e)}"

def call_gemini(api_key, user_prompt):
    """
    Hypothetical function for Gemini LLM.
    Replace with real Gemini logic.
    """
    return "(Gemini usage not yet implemented; placeholder)"

def call_llm(api_provider, api_key, user_prompt):
    """Routes calls to the correct LLM provider."""
    if not api_key:
        return "No API key provided. Using GPT-4o-mini default is not possible without HF key." if api_provider.lower() == "gpt-4o-mini" else "No API key provided."

    provider_lower = api_provider.lower()
    if provider_lower == "gpt-4o-mini":
        return call_gpt4o_mini(api_key, user_prompt)
    elif provider_lower == "openai":
        return call_openai(api_key, user_prompt)
    elif provider_lower == "deepseek":
        return call_deepseek(api_key, user_prompt)
    elif provider_lower == "gemini":
        return call_gemini(api_key, user_prompt)
    else:
        return f"Unknown provider: {api_provider}. Please choose GPT-4o-mini, OpenAI, DeepSeek, or Gemini."

# --------------------------------------------------------------------------------
# ADVANCED DATA ANALYSIS (extended_analysis) 
# --------------------------------------------------------------------------------
def extended_analysis(df):
    """
    Sample advanced analysis:
      1. Correlation heatmap for numeric columns
      2. Bar plot of 'Career' (if present)
      3. Simple logistic regression classification if 'Career' is suitable
    """
    output_paths = []
    numeric_cols = df.select_dtypes(include=["number"]).columns.tolist()

    cat_cols = df.select_dtypes(exclude=["number"]).columns.tolist()

    # 1) Correlation Heatmap
    if len(numeric_cols) > 1:
        corr = df[numeric_cols].corr()
        plt.figure(figsize=(8, 6))
        sns.heatmap(corr, annot=True, cmap="coolwarm", fmt=".2f")
        plt.title("Correlation Heatmap")
        heatmap_path = "heatmap.png"
        plt.savefig(heatmap_path)
        plt.close()
        output_paths.append(heatmap_path)

    # 2) Bar Plot of 'Career' if present
    if "Career" in df.columns:
        plt.figure(figsize=(8, 5))
        df["Career"].value_counts().plot(kind="bar")
        plt.title("Count of Each Career")
        plt.xlabel("Career")
        plt.ylabel("Count")
        barplot_path = "barplot_career.png"
        plt.savefig(barplot_path)
        plt.close()
        output_paths.append(barplot_path)

    # 3) Simple Logistic Regression if 'Career' exists with multiple categories
    if "Career" in df.columns and len(numeric_cols) > 0:
        le = LabelEncoder()
        df["Career_encoded"] = le.fit_transform(df["Career"])
        X = df[numeric_cols].fillna(0)
        y = df["Career_encoded"]
        if len(np.unique(y)) > 1:
            X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
            model = LogisticRegression(max_iter=1000)
            model.fit(X_train, y_train)
            score = model.score(X_test, y_test)
            accuracy_info = f"Logistic Regression accuracy on test set: {score:.2f}"
        else:
            accuracy_info = "Career column has only one class; no classification performed."
    else:
        accuracy_info = "No 'Career' column or insufficient numeric data for classification."

    return output_paths, accuracy_info

# --------------------------------------------------------------------------------
# MAIN ANALYSIS AND VISUALIZATION FUNCTION
# --------------------------------------------------------------------------------
def analyze_and_visualize(
    file,
    message,
    history,
    api_provider,
    api_key
):
    """
    Loads CSV, gives a summary, calls LLM for suggestions if an API key is provided,
    does extended analysis if user requests ("sample analysis", "extended analysis", etc.),
    and returns results/plots in the chatbot.
    """
    try:
        # Load CSV
        df = pd.read_csv(file.name)
        numeric_cols = df.select_dtypes(include=["number"]).columns.tolist()
        categorical_cols = df.select_dtypes(exclude=["number"]).columns.tolist()

        # Basic info
        summary = (
            f"**File**: {file.name}\n"
            f"**Shape**: {df.shape[0]} rows, {df.shape[1]} columns\n"
            f"**Numerical Columns**: {', '.join(numeric_cols) if numeric_cols else 'None'}\n"
            f"**Categorical Columns**: {', '.join(categorical_cols) if categorical_cols else 'None'}\n"
        )

        # LLM suggestions
        llm_suggestions = ""
        if api_key:
            user_prompt = (
                f"Data Summary:\n{summary}\n\n"
                f"User question or request: {message}\n"
                f"Suggest advanced data analysis or steps if relevant."
            )
            llm_response = call_llm(api_provider, api_key, user_prompt)
            llm_suggestions = f"\n**LLM Suggestions**:\n{llm_response}\n"
        else:
            llm_suggestions = "\n(No LLM suggestions because no API key provided.)\n"

        # Always produce example histogram if there's at least one numeric column
        hist_path = None
        if numeric_cols:
            plt.figure(figsize=(6, 4))
            sns.histplot(df[numeric_cols[0]], kde=True)
            plt.title(f"Distribution of '{numeric_cols[0]}'")
            plt.tight_layout()
            hist_path = "temp_plot.png"
            plt.savefig(hist_path)
            plt.close()

        # Check if the user wants extended analysis
        trigger_phrases = ["sample analysis", "extended analysis", "advanced analysis", "run analysis"]
        analysis_paths = []
        accuracy_info = ""
        if any(phrase in message.lower() for phrase in trigger_phrases):
            analysis_paths, accuracy_info = extended_analysis(df)

        # Build final response text
        response_text = summary + llm_suggestions
        if accuracy_info:
            response_text += f"\n**ML Model Info**: {accuracy_info}\n"

        # Construct the final chatbot content
        chat_content = [(message, response_text)]
        if hist_path:
            chat_content.append((None, (hist_path,)))
        for path in analysis_paths:
            chat_content.append((None, (path,)))

        return history + chat_content

    except Exception as e:
        return history + [(message, f"Error: {str(e)}")]

# --------------------------------------------------------------------------------
# CREATING THE GRADIO APP
# --------------------------------------------------------------------------------
def create_demo():
    with gr.Blocks() as demo:
        gr.Markdown("# 🤖 GPT-4o-mini (Default) + Multi-Provider AI Data Analysis Assistant")
        gr.Markdown(
            """
            **Features**:
            - Default LLM: GPT-4o-mini on Hugging Face (requires HF API key).
            - Other providers: **OpenAI**, **DeepSeek**, **Gemini** (enter their respective API keys).
            - Upload CSV for data summary & histograms.
            - Type "sample analysis" or "extended analysis" to trigger correlation heatmaps, bar plots, and a simple logistic regression.
            """
        )

        with gr.Row():
            api_provider = gr.Dropdown(
                choices=["GPT-4o-mini", "OpenAI", "DeepSeek", "Gemini"],
                value="GPT-4o-mini",  # default
                label="LLM Provider",
            )
            api_key = gr.Textbox(
                label="LLM API Key",
                placeholder="Enter your Hugging Face/DeepSeek/OpenAI/Gemini API key here..."
            )

        file_input = gr.File(label="Upload CSV File", file_types=[".csv"])
        chatbot = gr.Chatbot(label="Analysis Output")
        msg = gr.Textbox(
            label="Message",
            placeholder="Ask the AI or type 'sample analysis' for extended analysis..."
        )

        send_btn = gr.Button("Send")
        reset_btn = gr.Button("Reset Chat")

        def reset_chat():
            return []

        msg.submit(
            fn=lambda f, m, h, p, k: analyze_and_visualize(f, m, h or [], p, k),
            inputs=[file_input, msg, chatbot, api_provider, api_key],
            outputs=[chatbot]
        ).then(lambda: "", None, [msg])

        send_btn.click(
            fn=lambda f, m, h, p, k: analyze_and_visualize(f, m, h or [], p, k),
            inputs=[file_input, msg, chatbot, api_provider, api_key],
            outputs=[chatbot]
        ).then(lambda: "", None, [msg])

        reset_btn.click(fn=reset_chat, inputs=[], outputs=[chatbot])

        demo.queue()
        return demo


demo = create_demo()

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