import streamlit as st
import numpy as np
import joblib
import os
import matplotlib.pyplot as plt
import pandas as pd
from sklearn.metrics import (
    ConfusionMatrixDisplay,
    RocCurveDisplay,
    roc_auc_score
)

st.set_page_config(page_title="Credit Card Fraud Detection", layout="centered")

BASE_DIR = os.path.dirname(__file__)
MODEL_PATH = os.path.join(BASE_DIR, "model.pkl")
model = joblib.load(MODEL_PATH)

# ================= LOAD TEST DATA =================
X_test_path = os.path.join(BASE_DIR, "X_test.npy")
y_test_path = os.path.join(BASE_DIR, "y_test.npy")

if os.path.exists(X_test_path):
    X_test = np.load(X_test_path)
    y_test = np.load(y_test_path)

    y_prob = model.predict_proba(X_test)[:, 1]
    y_pred = model.predict(X_test)

    auc = roc_auc_score(y_test, y_prob)
else:
    X_test, y_test, y_prob, y_pred, auc = None, None, None, None, None

# ================= TITLE =================
st.title("💳 Credit Card Fraud Detection")
st.markdown("Real-time fraud detection using Machine Learning.")

threshold = st.slider("Fraud detection threshold", 0.0, 1.0, 0.5)

# ================= TABS =================
tab1, tab2 = st.tabs(["🔍 Prediction", "📊 Model Insights"])

# ================= TAB 1 =================
with tab1:

    st.subheader("Transaction Input")

    if st.button("🎲 Generate Random Transaction"):
        time = np.random.randint(0, 172800)
        amount = np.random.uniform(1, 5000)
        pca_values = np.random.normal(0, 1, 28)

    else:
        col1, col2 = st.columns(2)

        with col1:
            time = st.number_input("Time", value=0)

        with col2:
            amount = st.number_input("Amount", value=0.0)

        st.info("V1–V28 are PCA-transformed anonymized features.")

        pca_values = []
        with st.expander("PCA Features"):
            for i in range(28):
                pca_values.append(st.number_input(f"V{i+1}", value=0.0))

    features = np.array([[time] + list(pca_values) + [amount]])

    if st.button("Predict"):

        with st.spinner("Analyzing transaction..."):
            prob = model.predict_proba(features)[0][1]
            pred = int(prob >= threshold)

        st.markdown("## Result")

        col1, col2 = st.columns(2)

        with col1:
            if pred == 1:
                st.error("🚨 Fraudulent Transaction")
            else:
                st.success("✅ Legitimate Transaction")

        with col2:
            st.metric("Fraud Probability", f"{prob:.4f}")

        st.progress(prob)

        # Download report
        report = pd.DataFrame({
            "Time": [time],
            "Amount": [amount],
            "Fraud Probability": [prob],
            "Prediction": [pred]
        })

        st.download_button(
            "📥 Download prediction report",
            report.to_csv(index=False),
            "prediction.csv"
        )

# ================= TAB 2 =================
with tab2:

    st.subheader("Model Performance")

    if X_test is None:
        st.info("Upload X_test.npy and y_test.npy to enable insights.")

    else:
        st.metric("AUC", round(auc, 3))

        fig, ax = plt.subplots()
        RocCurveDisplay.from_predictions(y_test, y_prob, ax=ax)
        st.pyplot(fig)

        fig2, ax2 = plt.subplots()
        ConfusionMatrixDisplay.from_predictions(y_test, y_pred, ax=ax2)
        st.pyplot(fig2)

    if hasattr(model, "feature_importances_"):

        st.subheader("Feature Importance")

        fig3, ax3 = plt.subplots()
        ax3.bar(range(len(model.feature_importances_)), model.feature_importances_)
        ax3.set_title("Feature Importance")
        st.pyplot(fig3)