import gradio as gr
from transformers import pipeline
import pandas as pd
from datetime import datetime
import os

# -----------------------------
# Load pretrained emotion model
# -----------------------------
model_name = "superb/hubert-large-superb-er"
emotion_classifier = pipeline("audio-classification", model=model_name)

# Emotion mapping (short label → full name + emoji)
EMOTION_MAP = {
    "ang": ("Angry", "😡"),
    "hap": ("Happy", "😄"),
    "neu": ("Neutral", "😐"),
    "sad": ("Sad", "😢"),
    "exc": ("Excited", "🤩"),
    "fru": ("Frustrated", "😤"),
    "fea": ("Fearful", "😨"),
    "sur": ("Surprised", "😲"),
    "dis": ("Disgusted", "🤢"),
}

# -----------------------------
# Setup data storage
# -----------------------------
os.makedirs("data", exist_ok=True)
CSV_PATH = "data/customer_complaints_emotion.csv"

# Example categories — these can be linked to your delivery/order system
ORDER_ISSUES = [
    "Late Delivery",
    "Damaged Package",
    "Wrong Product Delivered",
    "Missing Items",
    "Rude Delivery Staff",
    "Payment Issue",
    "Return/Refund Request",
    "Order Cancelled Automatically",
    "Other Complaint"
]

# -----------------------------
# Emotion analysis function
# -----------------------------
def analyze_complaint(audio_file, complaint_type, order_id):
    if audio_file is None or not complaint_type or not order_id:
        return "⚠️ Please upload an audio complaint, select issue type, and enter Order ID."

    # Run prediction
    results = emotion_classifier(audio_file)
    results = sorted(results, key=lambda x: x['score'], reverse=True)
    top = results[0]
    label, emoji = EMOTION_MAP.get(top['label'], (top['label'], "🎭"))
    score = round(top['score'] * 100, 2)

    # -----------------------------
    # Build dashboard text
    # -----------------------------
    dashboard = f"## 📞 Customer Complaint Emotion Dashboard\n"
    dashboard += f"### 🎯 Detected Emotion: {emoji} **{label.upper()} ({score}%)**\n\n"
    dashboard += f"### 📦 Complaint Type: **{complaint_type}**\n"
    dashboard += f"### 🧾 Order ID: **{order_id}**\n\n"
    dashboard += f"### 📊 Emotion Breakdown\n"

    for r in results:
        lbl, emo = EMOTION_MAP.get(r['label'], (r['label'], "🎭"))
        scr = round(r['score'] * 100, 2)
        bar = "█" * int(scr / 5)
        dashboard += f"{emo} **{lbl}**: {scr}%  \n{bar}\n\n"

    # -----------------------------
    # Save results to CSV
    # -----------------------------
    df = pd.DataFrame({
        "datetime": [datetime.now().strftime("%Y-%m-%d %H:%M:%S")],
        "order_id": [order_id],
        "complaint_type": [complaint_type],
        "emotion": [label],
        "score": [score]
    })
    if os.path.exists(CSV_PATH):
        df.to_csv(CSV_PATH, mode="a", header=False, index=False)
    else:
        df.to_csv(CSV_PATH, index=False)

    return dashboard


# -----------------------------
# Gradio Interface
# -----------------------------
app = gr.Interface(
    fn=analyze_complaint,
    inputs=[
        gr.Audio(type="filepath", label="🎙️ Upload Customer Complaint Voice"),
        gr.Dropdown(ORDER_ISSUES, label="🚚 Select Complaint Type"),
        gr.Textbox(label="📦 Enter Order ID"),
    ],
    outputs="markdown",
    title="🎧 Customer Service Emotion Analyzer",
    description=(
        "Analyze voice-based customer complaints to detect emotions and "
        "prioritize based on emotional intensity. Ideal for supply chain "
        "and logistics feedback monitoring."
    ),
    allow_flagging="never",
)

# -----------------------------
# Run the app
# -----------------------------
if __name__ == "__main__":
    app.launch()