Glosarium AI
Kamus lengkap Kecerdasan Buatan
Feature Engineering
Process of creating predictive variables from raw data through transformation, aggregation, and combination to improve machine learning model performance.
Feature Pipeline
Automated workflow that transforms raw data into ready-to-use features, including extraction, transformation, validation, and loading into the feature store.
Online Feature Store
Component of the feature store optimized for low-latency access, storing the most recent features for real-time predictions of production models.
Offline Feature Store
Repository of historical feature data optimized for analytical queries and model training, typically based on data lake or data warehouse technologies.
Feature Discovery
Process of exploring and identifying relevant features from raw or existing data, often assisted by automated analysis techniques.
Feature Registry
Centralized catalog that documents all available features with their metadata, definitions, statistics, and use cases to promote reusability.
Point-in-time Correctness
Guarantee that features used for training exactly represent information available at a specific point in time in the past, avoiding future data leakage.
Feature Monitoring
Continuous monitoring of feature distributions and quality in production to detect drifts, anomalies, and schema breaks that could affect model performance.
Feature Lineage
Complete traceability of the origin and transformations applied to each feature, from source data to its use in machine learning models.
Feature Versioning
Management of different versions of a feature over time, allowing exact reproduction of training conditions and management of progressive migrations in production.
Feature Transformation
Application of mathematical or statistical functions on raw features to normalize, standardize, or encode variables according to machine learning algorithm requirements.
Feature Store Backfilling
Process of retrospective computation and storage of historical features to ensure temporal consistency and enable training on complete data periods.
Feature Serving
Mechanism for delivering features to consumer applications, optimized for latency and scalability, whether for batch training or real-time predictions.
Feature Store Governance
Set of policies, procedures, and controls ensuring quality, security, compliance, and appropriate use of features within the organization.
Feature Store Architecture
System design of the feature store including storage, compute, service, and orchestration components to ensure scalability, performance, and consistency between training and production.
Feature Store as a Service
Managed feature store offering provided by cloud or specialized vendors, eliminating the need to deploy and maintain the underlying infrastructure.
Feature Store Latency
Measurement of the response time of the feature store between the feature request and their availability, critical for real-time prediction applications.