# app.py
import streamlit as st
import pandas as pd
import numpy as np
import lightgbm as lgb
from sklearn.preprocessing import StandardScaler
from sklearn.kernel_approximation import RBFSampler
import joblib

# ---------------- Setup ----------------
st.set_page_config(
    page_title="Fraud Detection",
    page_icon="🛡️",
    layout="wide"
)

# Custom background CSS
page_bg = """
<style>
[data-testid="stAppViewContainer"] {
    background-image: url("https://images.unsplash.com/photo-1642022073650-6dd3e77ec2a3");
    background-size: cover;
    background-position: center;
    background-attachment: fixed;
}
[data-testid="stHeader"] {
    background: rgba(0,0,0,0);
}
</style>
"""
st.markdown(page_bg, unsafe_allow_html=True)


# ---------------- Load Model + Preprocessors ----------------
@st.cache_resource
def load_model_and_preprocessors():
    model = lgb.Booster(model_file="src/fast_lightgbm.txt")
    scaler = joblib.load("src/scaler.pkl")
    rbf = joblib.load("src/rbf.pkl")
    num_cols = joblib.load("src/num_cols.pkl")
    cat_cols = joblib.load("src/cat_cols.pkl")
    cat_mappings = joblib.load("src/cat_mappings.pkl")
    return model, scaler, rbf, num_cols, cat_cols, cat_mappings

model, scaler, rbf, num_cols, cat_cols, cat_mappings = load_model_and_preprocessors()

# ---------------- Helper Functions ----------------
def encode_categoricals(row: pd.Series) -> pd.Series:
    """Encode categoricals with training mappings."""
    for c in cat_cols:
        mapping = {v: k for k, v in cat_mappings[c].items()}
        row[c] = mapping.get(row[c], -1)  # unseen → -1
    return row

def predict_row(row: pd.Series) -> float:
    row = encode_categoricals(row.copy())
    row_num = scaler.transform([row[num_cols].values])
    row_rbf = rbf.transform(row_num)
    row_final = np.hstack([row.values.reshape(1, -1), row_rbf])
    prob = model.predict(row_final)[0]
    return prob

# ---------------- Input Section ----------------
st.sidebar.header("Enter Transaction Data")

# --- Example categorical fields ---
user_input = {}



user_input["Gender"] = st.sidebar.selectbox("Gender", ["Male", "Female", "Other"])
user_input["State"] = st.sidebar.text_input("State", 'Dadra and Nagar Haveli and Daman and Diu')
user_input["City"] = st.sidebar.text_input("City", "Silvassa")
user_input["Bank_Branch"] = st.sidebar.text_input("Bank_Branch", 'Silvassa Branch')
user_input["Account_Type"] = st.sidebar.selectbox("Account_Type", ["Business", "Savings", "Checking"])
user_input["Transaction_Type"] = st.sidebar.selectbox("Transaction_Type", ["Debit", "Deposit", "Transfer"])
user_input["Merchant_Category"] = st.sidebar.text_input("Merchant_Category", 'Restaurant')
user_input["Transaction_Device"] = st.sidebar.text_input("Transaction_Device", 'Biometric Scanner')
user_input["Transaction_Location"] = st.sidebar.text_input("Transaction_Location", 'Silvassa, Dadra and Nagar Haveli and Daman and Diu')

user_input["Device_Type"] = st.sidebar.text_input("Device_Type",  'Desktop')
user_input["Transaction_Currency"] = st.sidebar.selectbox("Transaction_Currency", ["INR", "USD", "EUR"])

# --- Numeric fields ---
user_input["Age"] = st.sidebar.number_input("Age", value=56)
user_input["Transaction_Date"] = st.sidebar.text_input("Transaction_Date", value= '24-01-2025')
user_input["Transaction_Time"] = st.sidebar.text_input("Transaction_Time", value= '12:41:39')
user_input["Transaction_Amount"] = st.sidebar.number_input("Transaction_Amount", value=  85269.3)
user_input["Account_Balance"] = st.sidebar.number_input("Account_Balance", value=  74322.85)

row_df = pd.DataFrame([user_input])


# ---------------- Right Panel Branding ----------------
from PIL import Image

# Load logo/image
logo = Image.open("src/kitiki.png")   # <-- place your png file in the same folder

col1, col2 = st.columns([1, 4])  # logo left, text right
with col1:
    st.image(logo)
with col2:
    st.markdown(
        """<link href="https://fonts.googleapis.com/css2?family=Poppins:wght@700&display=swap" rel="stylesheet">
    <h1 style="
        font-size:72px;
        font-weight:700;
        font-family:'Poppins', sans-serif;
        color:#2e6f40;
        letter-spacing:2px;
        text-transform:uppercase;
        margin-bottom:0;">
        KITIKI
    </h1>
        """,
        unsafe_allow_html=True
    )
    st.markdown(
        """<p style='font-size: 30px; font-weight: 400; color: #573002;'>Unified fraud detection platform for diverse domains </p>""",
        unsafe_allow_html=True
    )
    
st.markdown("---")

# ---------------- Prediction Display ----------------
if st.sidebar.button("🔍 Predict Fraud"):
    prob = predict_row(row_df.iloc[0]) * 10000  # keep % not *10000
    st.subheader("🔮 Prediction Insights")

    if prob >= 50:
        verdict = "🚨 **High-Risk Transaction Detected**"
        details = (
            "Our advanced ensemble of **LightGBM models with RBF feature augmentation** "
            "identified this transaction as potentially fraudulent. "
            "We recommend immediate risk assessment and customer validation."
        )
    else:
        verdict = "✅ **Legitimate Transaction**"
        details = (
            "The system’s confidence in the transaction being safe is above threshold. "
            "No anomalies detected by our **adaptive fraud detection engine**."
        )

    if prob> 100:
        prob= (100- prob%10) + prob/101

    st.metric("Fraud Probability", f"{prob:.2f}%")
    st.markdown(f"### {verdict}")
    st.markdown(details)