Back to the Index

On-Demand Features

What are on-demand features?

‍If a feature is used in an online inference pipeline and it is created using data only available at request-time, then it is an on-demand feature. If you have historical data, you would like to be able to create the same feature values in your feature pipeline using the same code as is used in the online inference pipeline (to avoid online-offline skew).

What is the use case for on-demand features?

On-demand features are typically used when real-time data is critical to the machine learning model's performance or when data availability constraints require the feature to be calculated at request-time.

‍When should you use on-demand features?

‍In general, it is preferable to precompute features in feature pipelines and store them in a feature store, retrieving them when needed in your online inference pipeline. Having a single place (the feature pipeline) where features are computed prevents online-offline skew, and enables materialized feature values to be shared across multiple models, promoting collaboration and reducing duplication of effort. By centralizing and precomputing features, teams can maintain a single source of truth for features, which simplifies the development process and fosters consistency across projects. On-demand can lead to increased latency or computational overhead in the online inference pipeline, negatively impacting prediction latency. However, if the input data used to compute a feature is only available at request time, then you will need to implement an on-demand feature. For example, if you are predicting credit-card fraud online, some credit card transaction features (computed from the transaction purchase amount, category of purchase, location of purchase, timestamp for the purchase) will only be available as on-demand features.

Interested for more?

Does this content look outdated? If you are interested in helping us maintain this, feel free to contact us.
O
Auto-regressive Models
AutoML
O
Backfill features
Backfill training data
Backpressure for feature stores
Batch Inference Pipeline
O
CI/CD for MLOps
Compound AI Systems
Context Window for LLMs
O
DAG Processing Model
Data Compatibility
Data Contract
Data Lakehouse
Data Leakage
Data Modeling
Data Partitioning
Data Pipelines
Data Quality
Data Transformation
Data Type (for features)
Data Validation (for features)
Data-Centric ML
Dimensional Modeling and Feature Stores
Downstream
O
ELT
ETL
Embedding
Encoding (for Features)
Entity
O
Feature
Feature Data
Feature Engineering
Feature Freshness
Feature Function
Feature Groups
Feature Logic
Feature Monitoring
Feature Pipeline
Feature Platform
Feature Reuse
Feature Selection
Feature Service
Feature Store
Feature Type
Feature Value
Feature Vector
Feature View
Filtering
Fine-Tuning LLMs
O
Generative AI
Gradient Accumulation
O
Hallucinations in LLMs
Hyperparameter
Hyperparameter Tuning
O
Idempotent Machine Learning Pipelines
In Context Learning (ICL)
Inference Data
Inference Logs
Inference Pipeline
Instruction Datasets for Fine-Tuning LLMs
O
LLM Code Interpreter
LLMOps
LLMs - Large Language Models
Lagged features
LangChain
Latent Space
O
ML
ML Artifacts (ML Assets)
MLOps
MVPS
Machine Learning Infrastructure
Machine Learning Logs
Machine Learning Observability
Machine Learning Pipeline
Machine Learning Systems
Model Architecture
Model Bias
Model Deployment
Model Development
Model Evaluation (Model Validation)
Model Governance
Model Inference
Model Interpretability
Model Monitoring
Model Performance
Model Quantization
O
Natural Language Processing (NLP)
O
Offline Store
On-Demand Transformation
Online Inference Pipeline
Online Store
Online-Offline Feature Skew
Online-Offline Feature Store Consistency
Orchestration
O
KServe
PagedAttention
Pandas UDF
Parameter-Efficient Fine-Tuning (PEFT) of LLMs
Point-in-Time Correct Joins
Precomputed Features
Prompt Engineering
Prompt Store
Prompt Tuning
Python UDF
O
RLHF - Reinforcement Learning from Human Feedback
Real-Time Machine Learning
Representation Learning
Retrieval Augmented Generation (RAG) for LLMs
RoPE Scaling
O
SQL UDF in Python
Sample Packing
Schema
Similarity Search
Skew
Splitting Training Data
Streaming Feature Pipeline
Streaming Inference Pipeline
O
Test Set
Theory-of-Mind Tasks
Time travel (for features)
Train (Training) Set
Training Data
Training Pipeline
Training-Inference Skew
Transformation
Two-Tower Embedding Model
Types of Machine Learning
O
Upstream
O
Validation Set
Vector Database
Versioning (of ML Artifacts)
The Feature StoreProduct CapabilitiesOpen SourceCustomersIntegrationsApp Status
The MLOps DictionaryEU AI Act GuideExamplesUse-CasesBlogEventsDocumentationFeature Store ComparisonCommunityFAQ
About UsContact UsSecurity & Compliance
Slack
Github
Twitter
Linkedin
Youtube

© Hopsworks 2024. All rights reserved. Various trademarks held by their respective owners.

Privacy Policy
Cookie Policy
Terms and Conditions