Feature Engineering & Feature Stores – Fueling ML with Quality Features

Key Learnings

• What feature engineering is and why it’s crucial for ML success.
• Types of features and common transformations.
• Challenges in feature consistency across train vs inference.
• What feature stores are and how they help.
• Popular feature stores overview: Feast, Tecton, SageMaker Feature Store.

Learn: What is Feature Engineering?

Transform raw data into meaningful input features that improve model performance and generalization.

− 100% +

Practical Example

Feature engineering on house_prices.csv including house age, size per room, one‑hot location, and scaling.

import pandas as pd
from sklearn.preprocessing import StandardScaler

# Load CSV
df = pd.read_csv('house_prices.csv')

# Create features
current_year = 2025
df['house_age'] = current_year - df['built_year']
df['size_per_room'] = df['size_sqft'] / df['bedrooms']

# One-hot encode 'location'
df = pd.get_dummies(df, columns=['location'], prefix='location')

# Normalize numerical features
scaler = StandardScaler()
df[['size_sqft', 'house_age', 'size_per_room']] = scaler.fit_transform(
    df[['size_sqft', 'house_age', 'size_per_room']]
)

# Drop unused columns
df = df.drop(['built_year'], axis=1)

# Save
df.to_csv('house_prices_engineered.csv', index=False)
print("🧠 Final Feature Engineered DataFrame:")
print(df.head())

Common Feature Types

• Numerical, Categorical, Ordinal, Binary
• Datetime, Text/NLP, Boolean
• Geospatial, Image/Audio, Sensor/IoT

Transforms

• Encoding, Scaling, Binning, Datetime extraction
• NLP tokenization, Log transforms, Interactions
• Imputation, Polynomial features, Quantiles

Feast (Open Source)

Open-source feature store supporting batch and real‑time, with online/offline stores and a pluggable backend.

pip install feast
feast init feast_project
cd feast_project

# Define a FileSource and FeatureView (example)
from datetime import timedelta
from feast import Entity, FeatureView, Field, FileSource
from feast.types import Int64, Float64

engagement_source = FileSource(
    path="customer_engagement.csv",
    timestamp_field="signup_date"
)

customer = Entity(name="customer_id", join_keys=["customer_id"])

engagement_fv = FeatureView(
    name="engagement_fv",
    entities=["customer_id"],
    ttl=timedelta(days=365),
    schema=[
        Field(name="last_login_days", dtype=Int64),
        Field(name="num_sessions", dtype=Int64),
        Field(name="avg_session_duration", dtype=Float64),
    ],
    source=engagement_source
)

Tecton (Managed)

Enterprise managed feature store with declarative pipelines, lineage, monitoring, and both streaming/batch support.

• Automated transformations, versioning, governance
• Integrates with Snowflake, Spark, Kafka

tecton.ai

SageMaker Feature Store

Fully managed AWS feature store with deep integration into SageMaker, IAM, encryption, and online/offline sync.

• Use within AWS‑native ML workflows
• CloudWatch observability and Glue/Athena integration

AWS SageMaker Feature Store