src/streamlit_app.py

import os
import pandas as pd
import numpy as np
import streamlit as st
from dotenv import load_dotenv
from huggingface_hub import InferenceClient, login
import google.generativeai as genai
from io import StringIO
import time
import requests

# ======================================================
# ⚙️ APP CONFIGURATION
# ======================================================
st.set_page_config(page_title="📊 Smart Data Analyst Pro", layout="wide")
st.title("📊 Smart Data Analyst Pro (Chat Mode)")
st.caption("Chat with your dataset — AI cleans, analyzes, and visualizes data. Hugging Face + Gemini compatible.")

# ======================================================
# 🔐 Load Environment Variables
# ======================================================
load_dotenv()
HF_TOKEN = os.getenv("HF_TOKEN") or os.getenv("HUGGINGFACE_API_KEY")
GEMINI_API_KEY = os.getenv("GEMINI_API_KEY")

if not HF_TOKEN:
    st.error("❌ Missing HF_TOKEN. Please set it in your .env file.")
else:
    login(token=HF_TOKEN)

if GEMINI_API_KEY:
    genai.configure(api_key=GEMINI_API_KEY)
else:
    st.warning("⚠️ Gemini API key missing. Gemini 2.5 Flash will not work.")

# ======================================================
# 🧠 MODEL SETUP
# ======================================================
with st.sidebar:
    st.header("⚙️ Model Settings")

    CLEANER_MODEL = st.selectbox(
        "Select Cleaner Model:",
        [
            "Qwen/Qwen2.5-Coder-14B",
            "mistralai/Mistral-7B-Instruct-v0.3"
        ],
        index=0
    )

    ANALYST_MODEL = st.selectbox(
        "Select Analysis Model:",
        [
            "Gemini 2.5 Flash (Google)",
            "Qwen/Qwen2.5-14B-Instruct",
            "mistralai/Mistral-7B-Instruct-v0.3",
            "HuggingFaceH4/zephyr-7b-beta"
        ],
        index=0
    )

    temperature = st.slider("Temperature", 0.0, 1.0, 0.3)
    max_tokens = st.slider("Max Tokens", 128, 4096, 1024)

hf_cleaner_client = InferenceClient(model=CLEANER_MODEL, token=HF_TOKEN)
hf_analyst_client = None
if ANALYST_MODEL != "Gemini 2.5 Flash (Google)":
    hf_analyst_client = InferenceClient(model=ANALYST_MODEL, token=HF_TOKEN)

# ======================================================
# 🧩 SAFE GENERATION FUNCTION
# ======================================================
def safe_hf_generate(client, prompt, temperature=0.3, max_tokens=512, retries=2):
    """Try text generation, with retry + fallback on service errors."""
    for attempt in range(retries + 1):
        try:
            resp = client.text_generation(
                prompt,
                temperature=temperature,
                max_new_tokens=max_tokens,
                return_full_text=False,
            )
            return resp.strip()
        except Exception as e:
            err = str(e)
            # 🩹 FIX: Handle common server overloads gracefully
            if "503" in err or "Service Temporarily Unavailable" in err:
                time.sleep(2)
                if attempt < retries:
                    continue  # retry
                else:
                    return "⚠️ The Hugging Face model is temporarily unavailable. Please try again or switch to Gemini."
            elif "Supported task: conversational" in err:
                chat_resp = client.chat_completion(
                    messages=[{"role": "user", "content": prompt}],
                    max_tokens=max_tokens,
                    temperature=temperature,
                )
                return chat_resp["choices"][0]["message"]["content"].strip()
            else:
                raise e
    return "⚠️ Failed after retries."

# ======================================================
# 🧩 DATA CLEANING
# ======================================================
def fallback_clean(df: pd.DataFrame) -> pd.DataFrame:
    df = df.copy()
    df.dropna(axis=1, how="all", inplace=True)
    df.columns = [c.strip().replace(" ", "_").lower() for c in df.columns]
    for col in df.columns:
        if df[col].dtype == "O":
            if not df[col].mode().empty:
                df[col].fillna(df[col].mode()[0], inplace=True)
            else:
                df[col].fillna("Unknown", inplace=True)
        else:
            df[col].fillna(df[col].median(), inplace=True)
    df.drop_duplicates(inplace=True)
    return df

def ai_clean_dataset(df: pd.DataFrame) -> (pd.DataFrame, str):
    if len(df) > 50:
        return df, "⚠️ AI cleaning skipped: dataset has more than 50 rows."
    csv_text = df.to_csv(index=False)
    prompt = f"""
You are a professional data cleaning assistant.
Clean and standardize the dataset below dynamically:
1. Handle missing values
2. Fix column name inconsistencies
3. Convert data types (dates, numbers, categories)
4. Remove irrelevant or duplicate rows
Return ONLY a valid CSV text (no markdown, no explanations).

Dataset:
{csv_text}
"""
    try:
        cleaned_str = safe_hf_generate(hf_cleaner_client, prompt, temperature=0.1, max_tokens=4096)
        cleaned_str = cleaned_str.replace("```csv", "").replace("```", "").replace("###", "").strip()
        cleaned_df = pd.read_csv(StringIO(cleaned_str), on_bad_lines="skip")
        cleaned_df.columns = [c.strip().replace(" ", "_").lower() for c in cleaned_df.columns]
        return cleaned_df, "✅ AI cleaning completed successfully."
    except Exception as e:
        return df, f"⚠️ AI cleaning failed: {str(e)}"

# ======================================================
# 🧩 DATA SUMMARY (Token-efficient)
# ======================================================
def summarize_for_analysis(df: pd.DataFrame, sample_rows=10) -> str:
    summary = [f"Rows: {len(df)}, Columns: {len(df.columns)}"]
    for col in df.columns:
        non_null = int(df[col].notnull().sum())
        if pd.api.types.is_numeric_dtype(df[col]):
            desc = df[col].describe().to_dict()
            summary.append(f"- {col}: mean={desc.get('mean', np.nan):.2f}, median={df[col].median():.2f}, non_null={non_null}")
        else:
            top = df[col].value_counts().head(3).to_dict()
            summary.append(f"- {col}: top_values={top}, non_null={non_null}")
    sample = df.head(sample_rows).to_csv(index=False)
    summary.append("--- Sample Data ---")
    summary.append(sample)
    return "\n".join(summary)

# ======================================================
# 🧠 ANALYSIS FUNCTION
# ======================================================
def query_analysis_model(df: pd.DataFrame, user_query: str, dataset_name: str) -> str:
    prompt_summary = summarize_for_analysis(df)
    prompt = f"""
You are a professional data analyst.
Analyze the dataset '{dataset_name}' and answer the user's question.

--- DATA SUMMARY ---
{prompt_summary}

--- USER QUESTION ---
{user_query}

Respond with:
1. Key insights and patterns
2. Quantitative findings
3. Notable relationships or anomalies
4. Data-driven recommendations
"""
    try:
        if ANALYST_MODEL == "Gemini 2.5 Flash (Google)":
            response = genai.GenerativeModel("gemini-2.5-flash").generate_content(
                prompt,
                generation_config={
                    "temperature": temperature,
                    "max_output_tokens": max_tokens
                }
            )
            return response.text if hasattr(response, "text") else "No valid text response."
        else:
            # 🩹 FIX: wrap in retry-aware generator
            result = safe_hf_generate(hf_analyst_client, prompt, temperature=temperature, max_tokens=max_tokens)
            # fallback to Gemini if Hugging Face failed entirely
            if "temporarily unavailable" in result.lower() and GEMINI_API_KEY:
                alt = genai.GenerativeModel("gemini-2.5-flash").generate_content(prompt)
                return f"🔄 Fallback to Gemini:\n\n{alt.text}"
            return result
    except Exception as e:
        # 🩹 FIX: fallback if server rejects or 5xx
        if "503" in str(e) and GEMINI_API_KEY:
            response = genai.GenerativeModel("gemini-2.5-flash").generate_content(prompt)
            return f"🔄 Fallback to Gemini due to 503 error:\n\n{response.text}"
        return f"⚠️ Analysis failed: {str(e)}"

# ======================================================
# 🚀 MAIN CHATBOT LOGIC
# ======================================================
uploaded = st.file_uploader("📎 Upload CSV or Excel file", type=["csv", "xlsx"])
if "messages" not in st.session_state:
    st.session_state.messages = []

if uploaded:
    df = pd.read_csv(uploaded) if uploaded.name.endswith(".csv") else pd.read_excel(uploaded)

    with st.spinner("🧼 Cleaning your dataset..."):
        cleaned_df, cleaning_status = ai_clean_dataset(df)

    st.subheader("✅ Cleaning Status")
    st.info(cleaning_status)
    st.subheader("📊 Dataset Preview")
    st.dataframe(cleaned_df.head(), use_container_width=True)

    st.subheader("💬 Chat with Your Dataset")
    for msg in st.session_state.messages:
        with st.chat_message(msg["role"]):
            st.markdown(msg["content"])

    if user_query := st.chat_input("Ask something about your dataset..."):
        st.session_state.messages.append({"role": "user", "content": user_query})
        with st.chat_message("user"):
            st.markdown(user_query)

        with st.chat_message("assistant"):
            with st.spinner("🤖 Analyzing..."):
                result = query_analysis_model(cleaned_df, user_query, uploaded.name)
                st.markdown(result)
                st.session_state.messages.append({"role": "assistant", "content": result})
else:
    st.info("📥 Upload a dataset to begin chatting with your AI analyst.")