from sklearn.decomposition import TruncatedSVD
import pandas as pd
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.neighbors import NearestNeighbors
from sklearn.cluster import KMeans
from sklearn.metrics.pairwise import cosine_similarity
import gradio as gr

def custom_similarity_by_index(idx1, idx2):
    row1 = df.iloc[idx1]
    row2 = df.iloc[idx2]

    def to_set(text):
        if pd.isna(text):
            return set()
        return set(text.lower().split())

    g1, g2 = to_set(row1["genres"]), to_set(row2["genres"])
    k1, k2 = to_set(row1["keywords"]), to_set(row2["keywords"])

    genre_sim = len(g1 & g2) / len(g1 | g2) if g1 and g2 else 0
    keyword_sim = len(k1 & k2) / len(k1 | k2) if k1 and k2 else 0

    d1 = str(row1["director"]).lower()
    d2 = str(row2["director"]).lower()
    director_sim = 1 if d1 == d2 else 0

    return 0.4 * genre_sim + 0.4 * keyword_sim + 0.2 * director_sim

# ------------------ LOAD DATA ------------------
df = pd.read_csv("mov_dataset.csv")
df["title_lower"] = df["title"].str.lower()

tfidf = TfidfVectorizer()
X = tfidf.fit_transform(df["combined_features"])

# ------------------ SVD ------------------
svd = TruncatedSVD(n_components=50, random_state=42)
X_svd = svd.fit_transform(X)

# ------------------ MODELS ------------------
knn = NearestNeighbors(metric='cosine', algorithm='brute')
knn.fit(X_svd)

kmeans = KMeans(n_clusters=10, random_state=42)
clusters = kmeans.fit_predict(X_svd)

cosine_sim = cosine_similarity(X_svd)

# ------------------ RECOMMEND FUNCTION ------------------
def recommend(movie_name, model_type):
    movie_name = movie_name.strip().lower()

    if not movie_name:
        return ["Please enter a movie name"]

    if movie_name not in df["title_lower"].values:
        return ["Movie not found"]

    idx = df[df["title_lower"] == movie_name].index[0]

    if model_type == "KNN":
        distances, indices = knn.kneighbors([X_svd[idx]], n_neighbors=6)
    
        results = []
        for i in indices[0][1:]:
            score = custom_similarity_by_index(idx, i)
            results.append((df.iloc[i]["title"], score))
    
        return sorted(results, key=lambda x: x[1], reverse=True)

    elif model_type == "Cosine":
        scores = list(enumerate(cosine_sim[idx]))
        scores = sorted(scores, key=lambda x: x[1], reverse=True)
    
        results = []
        for i, _ in scores[1:6]:
            score = custom_similarity_by_index(idx, i)
            results.append((df.iloc[i]["title"], score))
    
        return sorted(results, key=lambda x: x[1], reverse=True)

    elif model_type == "KMeans":
        cluster_id = clusters[idx]
        cluster_indices = [i for i in range(len(df)) if clusters[i] == cluster_id and i != idx]
    
        if not cluster_indices:
            return ["No similar movies found"]
    
        results = []
        for i in cluster_indices:
            score = custom_similarity_by_index(idx, i)
            results.append((df.iloc[i]["title"], score))
    
        results = sorted(results, key=lambda x: x[1], reverse=True)
    
        return results[:5]



# ------------------ CARD STYLE OUTPUT ------------------
def recommend_ui(movie, model):
    results = recommend(movie, model)

    if isinstance(results, list) and isinstance(results[0], str):
        return f"❌ {results[0]}"

    cards = ""
    for i, (title, score) in enumerate(results):
        cards += f"""
        <div style="
            background-color:#1f1f1f;
            padding:15px;
            margin:10px;
            border-radius:10px;
            box-shadow:0px 0px 10px rgba(255,0,0,0.3);
        ">
            <h3 style="color:#E50914;">{i+1}. {title}</h3>
            <p style="color:white;">⭐ Similarity Score: <b>{score}</b></p>
        </div>
        """

    return cards


# ------------------ NETFLIX STYLE UI ------------------
css = """
body {
    background-color: #141414;
    color: white;
}

.gradio-container {
    background-color: #141414 !important;
}

h1 {
    text-align: center;
    color: #E50914;
    font-size: 42px;
}

"""


with gr.Blocks(css=css) as demo:

    gr.Markdown("# 🎬 Netflix Movie Recommender")

    with gr.Row():
        movie_input = gr.Textbox(
            label="Enter Movie Name",
            placeholder="Try: Avatar"
        )

        model_input = gr.Dropdown(
            ["KNN", "Cosine", "KMeans"],
            value="KNN",
            label="Select Model"
        )

    btn = gr.Button("🔥 Recommend")

    output = gr.HTML()

    btn.click(fn=recommend_ui, inputs=[movie_input, model_input], outputs=output)

demo.launch()