Praktikum 6

1. Lakukan percobaan penggunaan ANNOY, FAISS, dan HNSWLIB pada dataset sekunder berukuran besar (Micro Spotify) pada link berikut: https://www.kaggle.com/datasets/bwandowando/spotify-songs-with-attributes-and-lyrics/data . Download data dan load CSV filenya (pilih dataset yg pertama dari dua dataset). pilih hanya fitur numerik saja, dan lakukan normalisasi menggunakan StandardScaler. Lakukan pencarian track terdekat dan bandingkan hasilnya.

import pandas as pd
import numpy as np
import time
import faiss
from annoy import AnnoyIndex
import hnswlib
from sklearn.neighbors import NearestNeighbors
from sklearn.preprocessing import StandardScaler

# -------------------------------
# Load dataset
# -------------------------------
df = pd.read_csv('/content/drive/MyDrive/Kuliah/2025 ML/spotify_songs.csv')  # ganti path sesuai lokasi file
features = ['danceability', 'energy', 'loudness', 'speechiness', 
            'acousticness', 'instrumentalness', 'liveness', 'valence', 'tempo']
X = df[features].values

# Standarisasi fitur
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)

k = 10  # jumlah nearest neighbors

# -------------------------------
# Exact Nearest Neighbor (brute-force)
# -------------------------------
start = time.time()
nn = NearestNeighbors(n_neighbors=k, algorithm='brute', metric='euclidean')
nn.fit(X_scaled)
dist_exact, idx_exact = nn.kneighbors(X_scaled)
time_exact = time.time() - start
print(f"Exact NN done in {time_exact:.3f} s")

# -------------------------------
# Annoy
# -------------------------------
start = time.time()
f = X_scaled.shape[1]
index_annoy = AnnoyIndex(f, 'euclidean')
for i, v in enumerate(X_scaled):
    index_annoy.add_item(i, v)
index_annoy.build(10)
idx_annoy = [index_annoy.get_nns_by_vector(v, k) for v in X_scaled]
time_annoy = time.time() - start
print(f"Annoy done in {time_annoy:.3f} s")

# -------------------------------
# HNSW
# -------------------------------
start = time.time()
p_hnsw = hnswlib.Index(space='l2', dim=X_scaled.shape[1])
p_hnsw.init_index(max_elements=X_scaled.shape[0], ef_construction=200, M=16)
p_hnsw.add_items(X_scaled)
p_hnsw.set_ef(200)
idx_hnsw, dist_hnsw = p_hnsw.knn_query(X_scaled, k=k)
time_hnsw = time.time() - start
print(f"HNSW done in {time_hnsw:.3f} s")

# -------------------------------
# FAISS IVF
# -------------------------------
start = time.time()
quantizer = faiss.IndexFlatL2(X_scaled.shape[1])
index_faiss = faiss.IndexIVFFlat(quantizer, X_scaled.shape[1], nlist=100, metric=faiss.METRIC_L2)
index_faiss.train(X_scaled)
index_faiss.add(X_scaled)
index_faiss.nprobe = 10
dist_faiss, idx_faiss = index_faiss.search(X_scaled, k)
time_faiss = time.time() - start
print(f"FAISS IVF done in {time_faiss:.3f} s")

# -------------------------------
# Contoh tampilkan top-5 neighbors dari item pertama
# -------------------------------
print("\nTop-5 neighbors for first song:")
print(f"Exact NN: {idx_exact[0][:5]}")
print(f"Annoy:    {idx_annoy[0][:5]}")
print(f"HNSW:     {idx_hnsw[0][:5]}")
print(f"FAISS:    {idx_faiss[0][:5]}")

Buat dan tuliskan analisa anda terhadap code diatas.