import gradio as gr
import pandas as pd
import joblib
import matplotlib.pyplot as plt
import seaborn as sns
import io

# Load trained model
model = joblib.load("anomaly_detector_rf_model.pkl")

# Features used during training
feature_cols = [
    "amount", "hour", "day_of_week", "is_weekend", "merchant_avg_amount",
    "amount_zscore", "log_amount", "type_atm_withdrawal", "type_credit",
    "type_debit", "merchant_encoded"
]

# Function to detect anomalies
def detect_anomalies(df):
    original_df = df.copy()

    for col in ["transaction_id", "merchant", "location", "amount"]:
        if col not in original_df.columns:
            original_df[col] = "N/A" if col != "amount" else 0.0

    model_input = df.reindex(columns=feature_cols, fill_value=0)
    preds = model.predict(model_input)
    original_df["is_anomalous"] = preds

    anomalies = original_df[original_df["is_anomalous"] == 1]
    return original_df, anomalies[["transaction_id", "merchant", "location", "amount", "is_anomalous"]]

# Function to plot charts
def plot_charts(df):
    fig, axes = plt.subplots(2, 2, figsize=(12, 10))

    if "amount" in df.columns:
        sns.histplot(df["amount"], bins=30, kde=True, ax=axes[0, 0])
        axes[0, 0].set_title("Amount Distribution")
        sns.boxplot(x=df["amount"], ax=axes[0, 1])
        axes[0, 1].set_title("Amount Box Plot")
    else:
        axes[0, 0].text(0.5, 0.5, "No 'amount' column", ha='center')
        axes[0, 1].text(0.5, 0.5, "No 'amount' column", ha='center')

    if "day_of_week" in df.columns:
        sns.countplot(x=df["day_of_week"], ax=axes[1, 0])
        axes[1, 0].set_title("Transactions by Day of Week")
    else:
        axes[1, 0].text(0.5, 0.5, "No 'day_of_week' column", ha='center')

    if "merchant" in df.columns:
        top_merchants = df.groupby("merchant")["amount"].sum().nlargest(5).reset_index()
        sns.barplot(data=top_merchants, x="merchant", y="amount", ax=axes[1, 1])
        axes[1, 1].set_title("Top 5 Merchants by Amount")
    else:
        axes[1, 1].text(0.5, 0.5, "No 'merchant' column", ha='center')

    plt.tight_layout()
    return fig

# Function to generate summary + charts + file
def app_interface(csv_file):
    df = pd.read_csv(csv_file)
    full_df, anomalies = detect_anomalies(df)

    total = len(full_df)
    anom_count = len(anomalies)
    percent = (anom_count / total) * 100 if total > 0 else 0

    summary = (
        f"🔢 **Total Transactions**: {total}\n"
        f"⚠️ **Anomalies Detected**: {anom_count}\n"
        f"📊 **Anomaly Percentage**: {percent:.2f}%"
    )

    # Convert anomalies to CSV bytes for download
    csv_bytes = anomalies.to_csv(index=False).encode()
    download = io.BytesIO(csv_bytes)

    fig = plot_charts(full_df)

    return summary, anomalies, fig, download

# Gradio App with UI
with gr.Blocks(theme=gr.themes.Soft()) as interface:
    gr.Markdown("# 🛡️ Financial Abuse & Anomaly Detection App")
    gr.Markdown("Upload your **transaction CSV** to detect anomalies and view insights.")

    with gr.Row():
        file_input = gr.File(label="📁 Upload CSV File", file_types=[".csv"])
        detect_button = gr.Button("🚨 Run Detection", variant="primary")

    with gr.Row():
        summary_box = gr.Markdown("")

    with gr.Tab("📋 Anomalies Detected"):
        result_table = gr.Dataframe(label="🔴 Anomalies")
        download_btn = gr.File(label="⬇️ Download Detected Anomalies")

    with gr.Tab("📊 Transaction Charts"):
        chart_output = gr.Plot()

    detect_button.click(fn=app_interface, inputs=file_input,
                        outputs=[summary_box, result_table, chart_output, download_btn])

interface.launch(share=True)