# ✅ app.py - InsightForge AI with Full Preprocessing Control

import streamlit as st
import pandas as pd
from utils import (
    load_data, preprocess_data, train_models, train_regressors,
    visualize_results, explain_model, detect_problem_type
)
from pdf_report import generate_pdf

st.set_page_config(page_title="InsightForge AI", layout="wide")
st.title("📊 InsightForge AI - Business Intelligence AutoML Tool")

with st.sidebar:
    st.header("📁 Upload Your CSV")
    uploaded_file = st.file_uploader("Choose a file", type=["csv"])
    explain_mode = st.checkbox("🧠 Explain Like I'm 5 Mode (ELI5)")
    user_problem_type = st.radio("🔍 Select Problem Type", ["Auto Detect", "Regression", "Classification"])
    encoding_strategy = st.selectbox("🔤 Select Encoding Strategy", ["label", "onehot", "ordinal"])

if uploaded_file:
    df, error = load_data(uploaded_file)

    if error:
        st.error(f"❌ {error}")
    else:
        st.subheader("🔍 1. Preview of Raw Data")
        st.dataframe(df.head())

        if st.checkbox("🔀 Show Raw vs Clean Toggle"):
            st.write("📄 Raw Data:")
            st.dataframe(df.head())

        st.subheader("🛠️ 2. Preprocessing Summary")

        detected_type = detect_problem_type(df)
        problem_type = user_problem_type.lower() if user_problem_type != "Auto Detect" else detected_type

        df_clean, preprocess_summary = preprocess_data(df, explain_mode=explain_mode, problem_type=problem_type, encoding_strategy=encoding_strategy)
        st.dataframe(preprocess_summary)

        if st.checkbox("✅ Show Processed Data"):
            st.write("📊 Processed Data:")
            st.dataframe(df_clean.head())

        st.subheader("🌡️ 3. Correlation Heatmap")
        visualize_results(df_clean, stage="correlation")

        st.subheader("🤖 4. Modeling Results")

        X = df_clean.drop(columns=[df_clean.columns[-1]])
        y = df_clean[df_clean.columns[-1]]

        if problem_type == "regression":
            model_results = train_regressors(X, y)
            metric_key = "R2 Score"
        else:
            model_results = train_models(X, y)
            metric_key = "Accuracy"

        best_model_name = max(model_results, key=lambda x: model_results[x]["metrics"].get(metric_key, 0))
        best_model = model_results[best_model_name]["model"]

        st.dataframe(pd.DataFrame({k: v["metrics"] for k, v in model_results.items()}).T.round(3))
        st.success(f"🏆 Best Model: {best_model_name}")

        visualize_results(model_results, stage="model_eval", problem_type=problem_type)

        st.subheader("🧠 5. Explainable AI - Feature Importance")
        if explain_mode:
            if hasattr(best_model, 'feature_importances_'):
                explain_model(best_model, df_clean, problem_type)
            else:
                st.warning(
                    f"❌ The selected best model '{best_model_name}' does not support feature importance.\n"
                    f"Please choose a supported model (like Decision Tree)."
                )
                explain_choice = st.selectbox("📌 Choose a model to explain:", options=list(model_results.keys()))
                chosen_model = model_results[explain_choice]["model"]
                if hasattr(chosen_model, 'feature_importances_'):
                    explain_model(chosen_model, df_clean, problem_type)
                else:
                    st.error(f"⚠️ The chosen model '{explain_choice}' also does not support feature importance.")
        else:
            st.info("💡 Enable 'Explain Like I'm 5 Mode (ELI5)' from the sidebar to view feature importance.")

        if st.button("📥 Download PDF Report"):
            report_df = pd.DataFrame({k: v["metrics"] for k, v in model_results.items()}).T.round(3)
            generate_pdf(df, report_df, best_model_name, problem_type)
            st.success("✅ PDF Report generated and downloaded successfully.")
else:
    st.info("📂 Please upload a CSV file to begin.")