Tutorial Lengkap Vertex AI Feature Store: Manajemen Fitur Terpusat

Vertex AI Feature Store adalah repositori terpusat untuk mengorganisir, menyimpan, dan menyajikan fitur ML. Memungkinkan penggunaan ulang fitur, mengurangi training-serving skew, dan menyediakan akses fitur yang konsisten antar tim.

Mengapa Feature Store?

• Fitur terpusat: Satu sumber kebenaran
• Penggunaan ulang fitur: Berbagi fitur antar model
• Serving latensi rendah: Pengambilan fitur online cepat
• Konsistensi: Fitur sama untuk training dan serving
• Time travel: Pencarian fitur point-in-time

Prerequisites

pip install google-cloud-aiplatform

gcloud auth login
gcloud config set project your-project-id

Setup

1. Inisialisasi Vertex AI

from google.cloud import aiplatform

aiplatform.init(project="your-project-id", location="us-central1")

2. Buat Feature Store

# Buat feature store
featurestore = aiplatform.Featurestore.create(

    featurestoreid="myfeaturestore",
    onlinestorefixednodecount=1
)

print(f"Feature store dibuat: {featurestore.resourcename}")

Entity Types

1. Buat Entity Type

# Buat customer entity type
customerentity = featurestore.createentitytype(
    entitytypeid="customer",
    description="Customer entity untuk prediksi churn"
)

Buat product entity type
productentity = featurestore.createentitytype(
    entitytypeid="product",
    description="Product entity untuk rekomendasi"
)

2. List Entity Types

entitytypes = featurestore.listentitytypes()
for et in entitytypes:

    print(f"{et.entitytypeid}: {et.description}")

# Buat features untuk customer entity
customerentity.createfeature(

    featureid="age",
    valuetype="INT64",

    description="Usia customer"
)

customerentity.createfeature(

    featureid="tenuremonths",

    valuetype="INT64",
    description="Bulan sebagai customer"
)

customerentity.createfeature(
    featureid="monthlycharges",
    valuetype="DOUBLE",

    description="Tagihan bulanan"
)

customerentity.createfeature(

    featureid="totalcharges",

    valuetype="DOUBLE",
    description="Total tagihan sampai saat ini"
)

customerentity.createfeature(
    featureid="contracttype",
    valuetype="STRING",

    description="Tipe kontrak"
)

# Buat multiple features sekaligus
featuresconfig = {
    "age": {"valuetype": "INT64", "description": "Usia customer"},

    "tenuremonths": {"valuetype": "INT64", "description": "Tenure dalam bulan"},

    "monthlycharges": {"valuetype": "DOUBLE", "description": "Tagihan bulanan"},

    "totalcharges": {"valuetype": "DOUBLE", "description": "Total tagihan"},

    "contracttype": {"valuetype": "STRING", "description": "Tipe kontrak"}

}

customerentity.batchcreatefeatures(featuresconfig)

# Ingest features dari BigQuery
customerentity.ingestfrombq(
    featureids=["age", "tenuremonths", "monthlycharges", "totalcharges"],
    featuretime="updatetime",
    bqsourceuri="bq://project.dataset.customerfeatures",

    entityidfield="customerid"
)

2. Ingest dari DataFrame

import pandas as pd
from datetime import datetime

Buat feature dataframe
df = pd.DataFrame({
    "customerid": ["C001", "C002", "C003"],

    "age": [25, 35, 45],
    "tenuremonths": [12, 24, 36],