import gradio as gr
import pandas as pd
import numpy as np
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import seaborn as sns
from vaderSentiment.vaderSentiment import SentimentIntensityAnalyzer
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import LabelEncoder
from statsmodels.tsa.arima.model import ARIMA
import warnings
warnings.filterwarnings('ignore')

# ─── DATA SETUP (runs once on app load) ───
np.random.seed(42)

n_listings = 500
neighbourhoods = ['Le Marais', 'Montmartre', 'Latin Quarter', 'Bastille',
                  'Belleville', 'Oberkampf', 'Saint-Germain', 'Pigalle',
                  'Batignolles', 'Menilmontant', 'Republique', 'Nation']
room_types = ['Entire home/apt', 'Private room', 'Shared room']

listings = pd.DataFrame({
    'listing_id': range(1, n_listings + 1),
    'neighbourhood': np.random.choice(neighbourhoods, n_listings),
    'room_type': np.random.choice(room_types, n_listings, p=[0.55, 0.38, 0.07]),
    'accommodates': np.random.choice([1,2,3,4,5,6], n_listings, p=[0.1,0.3,0.25,0.2,0.1,0.05]),
    'bedrooms': np.random.choice([0,1,2,3], n_listings, p=[0.15,0.5,0.25,0.1]),
    'minimum_nights': np.random.choice([1,2,3,5,7,30], n_listings, p=[0.3,0.25,0.2,0.1,0.1,0.05]),
    'number_of_reviews': np.random.poisson(40, n_listings),
    'reviews_per_month': np.round(np.random.exponential(2.5, n_listings), 2),
    'host_is_superhost': np.random.choice([0, 1], n_listings, p=[0.7, 0.3]),
    'instant_bookable': np.random.choice([0, 1], n_listings, p=[0.4, 0.6]),
})

base_prices = {'Entire home/apt': 120, 'Private room': 55, 'Shared room': 25}
premium = ['Le Marais', 'Saint-Germain', 'Latin Quarter', 'Montmartre']
listings['price'] = listings.apply(
    lambda r: base_prices[r['room_type']] * (1.3 if r['neighbourhood'] in premium else 1.0)
    * np.random.uniform(0.6, 1.6), axis=1).round(2)
listings['review_scores_rating'] = np.clip(
    np.where(listings['host_is_superhost'] == 1,
             np.random.normal(4.7, 0.2, n_listings),
             np.random.normal(4.3, 0.4, n_listings)), 3.0, 5.0).round(2)

# Generate reviews
review_templates = {
    'positive': ["Amazing location, very clean and the host was super responsive!",
                 "Perfect apartment for our stay. Walking distance to everything.",
                 "Loved the cozy atmosphere. Would definitely come back!",
                 "Great value for money. The neighborhood is lovely.",
                 "Exceeded expectations! Beautiful decor and comfortable bed."],
    'neutral': ["Decent place, a bit noisy at night but overall okay.",
                "Good location but the apartment was smaller than expected.",
                "It was fine for the price. Nothing special but clean enough."],
    'negative': ["Disappointed. The photos were misleading and it was dirty.",
                 "Would not recommend. Noisy neighbors and broken appliances.",
                 "Not worth the price at all. Bed was uncomfortable."]
}

reviews_list = []
for _ in range(5000):
    lid = np.random.choice(listings['listing_id'])
    rating = listings.loc[listings['listing_id'] == lid, 'review_scores_rating'].values[0]
    probs = [0.75, 0.2, 0.05] if rating >= 4.5 else ([0.5, 0.35, 0.15] if rating >= 4.0 else [0.25, 0.35, 0.4])
    cat = np.random.choice(['positive', 'neutral', 'negative'], p=probs)
    reviews_list.append({'listing_id': lid, 'comments': np.random.choice(review_templates[cat])})
reviews = pd.DataFrame(reviews_list)

# Sentiment
analyzer = SentimentIntensityAnalyzer()
reviews['sentiment'] = reviews['comments'].apply(lambda x: analyzer.polarity_scores(str(x))['compound'])
listing_sent = reviews.groupby('listing_id')['sentiment'].mean().reset_index()
listing_sent.columns = ['listing_id', 'avg_sentiment']
listings = listings.merge(listing_sent, on='listing_id', how='left').fillna(0)

# Train Random Forest
le_room = LabelEncoder()
le_hood = LabelEncoder()
listings['room_enc'] = le_room.fit_transform(listings['room_type'])
listings['hood_enc'] = le_hood.fit_transform(listings['neighbourhood'])
med_rating = listings['review_scores_rating'].median()
med_reviews = listings['reviews_per_month'].median()
listings['HighPerformer'] = ((listings['review_scores_rating'] >= med_rating) &
                              (listings['reviews_per_month'] >= med_reviews)).astype(int)

features = ['price','accommodates','bedrooms','minimum_nights','number_of_reviews',
            'host_is_superhost','instant_bookable','avg_sentiment','room_enc','hood_enc']
X = listings[features]
y = listings['HighPerformer']
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
rf = RandomForestClassifier(n_estimators=100, random_state=42, max_depth=10)
rf.fit(X_train, y_train)

# ─── APP FUNCTIONS ───

def neighbourhood_dashboard(neighbourhood):
    subset = listings[listings['neighbourhood'] == neighbourhood]
    n = len(subset)
    avg_price = subset['price'].mean()
    avg_sent = subset['avg_sentiment'].mean()
    avg_rating = subset['review_scores_rating'].mean()
    pct_superhost = subset['host_is_superhost'].mean() * 100
    hp_pct = subset['HighPerformer'].mean() * 100

    summary = f"""## {neighbourhood} Dashboard
| Metric | Value |
|--------|-------|
| Total Listings | {n} |
| Average Price | €{avg_price:.0f}/night |
| Average Sentiment | {avg_sent:.3f} |
| Average Rating | {avg_rating:.2f}/5.0 |
| Superhost % | {pct_superhost:.0f}% |
| High Performers | {hp_pct:.0f}% |

### Recommendation
{"**Premium neighbourhood** — prices above city average. Focus on maintaining quality to justify pricing." if avg_price > listings['price'].mean() else "**Value neighbourhood** — room to increase prices if sentiment stays positive."}
"""
    # Chart
    fig, axes = plt.subplots(1, 2, figsize=(12, 4))
    subset['room_type'].value_counts().plot(kind='bar', ax=axes[0], color=['#3498db','#e67e22','#27ae60'])
    axes[0].set_title(f'Room Types in {neighbourhood}')
    axes[0].set_xticklabels(axes[0].get_xticklabels(), rotation=30)
    axes[1].hist(subset['price'], bins=15, color='#3498db', edgecolor='white')
    axes[1].axvline(listings['price'].mean(), color='red', linestyle='--', label='City avg')
    axes[1].set_title(f'Price Distribution in {neighbourhood}')
    axes[1].set_xlabel('Price (€)')
    axes[1].legend()
    plt.tight_layout()
    return summary, fig

def feature_importance_chart():
    importances = pd.Series(rf.feature_importances_, index=features).sort_values(ascending=True)
    fig, ax = plt.subplots(figsize=(10, 6))
    importances.plot(kind='barh', ax=ax, color='#3498db', edgecolor='white')
    ax.set_xlabel('Importance Score')
    ax.set_title('What Makes an Airbnb Listing a High Performer?', fontsize=14, fontweight='bold')
    plt.tight_layout()
    accuracy = rf.score(X_test, y_test)
    report = f"**Model Accuracy: {accuracy*100:.1f}%**\n\nTop 3 features: {', '.join(importances.tail(3).index.tolist()[::-1])}"
    return report, fig

def price_forecast(neighbourhood):
    subset = listings[listings['neighbourhood'] == neighbourhood]
    base = subset['price'].mean()
    months = pd.date_range('2023-01-01', periods=24, freq='MS')
    trend = np.linspace(0, base * 0.15, 24)
    seasonality = base * 0.1 * np.sin(np.linspace(0, 4*np.pi, 24))
    noise = np.random.normal(0, base * 0.03, 24)
    ts = pd.Series(base + trend + seasonality + noise, index=months)

    model = ARIMA(ts, order=(1,1,1))
    fitted = model.fit()
    forecast = fitted.forecast(steps=6)
    forecast_idx = pd.date_range(months[-1] + pd.DateOffset(months=1), periods=6, freq='MS')

    fig, ax = plt.subplots(figsize=(12, 5))
    ax.plot(ts.index, ts.values, 'b-o', markersize=4, label='Historical', linewidth=1.5)
    ax.plot(forecast_idx, forecast.values, 'r--o', markersize=4, label='Forecast', linewidth=1.5)
    ax.fill_between(forecast_idx, forecast.values*0.92, forecast.values*1.08, alpha=0.2, color='red')
    ax.set_title(f'{neighbourhood} — 6-Month Price Forecast (ARIMA)', fontsize=14, fontweight='bold')
    ax.set_ylabel('Average Price (€)')
    ax.legend()
    ax.grid(True, alpha=0.3)
    plt.tight_layout()

    change = ((forecast.values[-1] - ts.values[-1]) / ts.values[-1]) * 100
    info = f"**Current avg: €{ts.values[-1]:.0f}** → **Forecasted: €{forecast.values[-1]:.0f}** ({change:+.1f}% over 6 months)"
    return info, fig

def sentiment_overview():
    fig, axes = plt.subplots(1, 2, figsize=(14, 5))
    hood_sent = listings.groupby('neighbourhood')['avg_sentiment'].mean().sort_values()
    colors = ['#e74c3c' if v < 0.3 else '#f39c12' if v < 0.37 else '#27ae60' for v in hood_sent]
    hood_sent.plot(kind='barh', ax=axes[0], color=colors)
    axes[0].set_title('Sentiment by Neighbourhood')
    axes[0].set_xlabel('Average Sentiment')
    sh = listings.groupby('host_is_superhost')['avg_sentiment'].mean()
    sh.index = ['Regular', 'Superhost']
    sh.plot(kind='bar', ax=axes[1], color=['#3498db', '#e67e22'])
    axes[1].set_title('Superhost Effect on Sentiment')
    axes[1].set_xticklabels(axes[1].get_xticklabels(), rotation=0)
    plt.tight_layout()
    best = hood_sent.idxmax()
    worst = hood_sent.idxmin()
    info = f"**Best:** {best} ({hood_sent.max():.3f}) | **Worst:** {worst} ({hood_sent.min():.3f}) | **Superhost boost:** +{sh['Superhost']-sh['Regular']:.3f}"
    return info, fig

# ─── BUILD APP ───
with gr.Blocks(title="Airbnb Pricing & Satisfaction Optimizer", theme=gr.themes.Soft()) as app:
    gr.Markdown("""# 🏠 Airbnb Pricing & Guest Satisfaction Optimizer
    *AI for Big Data Management — Group Project | ESCP Business School*
    
    Analyze listing performance across Paris neighbourhoods using sentiment analysis, 
    random forest classification, and ARIMA price forecasting.
    """)

    with gr.Tab("📊 Neighbourhood Dashboard"):
        gr.Markdown("Select a neighbourhood to see key metrics and pricing insights.")
        hood_input = gr.Dropdown(choices=sorted(neighbourhoods), value='Le Marais', label="Neighbourhood")
        hood_btn = gr.Button("Analyze", variant="primary")
        hood_summary = gr.Markdown()
        hood_chart = gr.Plot()
        hood_btn.click(neighbourhood_dashboard, inputs=hood_input, outputs=[hood_summary, hood_chart])

    with gr.Tab("🌲 Feature Importance"):
        gr.Markdown("Which features most predict whether a listing will be a high performer?")
        fi_btn = gr.Button("Show Feature Importance", variant="primary")
        fi_text = gr.Markdown()
        fi_chart = gr.Plot()
        fi_btn.click(feature_importance_chart, outputs=[fi_text, fi_chart])

    with gr.Tab("📈 Price Forecast"):
        gr.Markdown("ARIMA(1,1,1) forecast of average listing prices for the next 6 months.")
        fc_input = gr.Dropdown(choices=sorted(neighbourhoods), value='Le Marais', label="Neighbourhood")
        fc_btn = gr.Button("Forecast", variant="primary")
        fc_text = gr.Markdown()
        fc_chart = gr.Plot()
        fc_btn.click(price_forecast, inputs=fc_input, outputs=[fc_text, fc_chart])

    with gr.Tab("💬 Sentiment Analysis"):
        gr.Markdown("Overview of guest sentiment across all neighbourhoods and host types.")
        sa_btn = gr.Button("Show Sentiment Overview", variant="primary")
        sa_text = gr.Markdown()
        sa_chart = gr.Plot()
        sa_btn.click(sentiment_overview, outputs=[sa_text, sa_chart])

if __name__ == "__main__":
    app.launch()