Hopsworks
Hopsworks is a collaborative feature store and ML platform for batch and real-time data. Built around the industry's most advanced and modular feature store that provides seamless integration for existing pipelines and helps bring models to production faster.

What and why?
A feature store is a platform for managing features when building and deploying machine learning models. A Feature Store, like Hopsworks, is used for faster iteration and faster deployment of models to production. It is beneficial for scaling and moving real-time machine learning use cases with more ease as well as increasing productivity for data and AI teams.

Get familiar with concepts and terms
To get familiar with feature stores we advise you to first review the common terms and concepts commonly used in Machine Learning Operations (MLOps). 

• Definitions → https://www.hopsworks.ai/mlops-dictionary
• Documentation → https://docs.hopsworks.ai/latest
• Deep dives and other content → https://www.hopsworks.ai/blog

Disclaimer
Hopsworks can be used in three different ways ; 

• Serverless; infrastructure-free method which allows everyone to use Hopsworks for free
• Managed Hopsworks; deployed in the user’s cloud infrastructure
• On-premises, for restricted deployment in the user’s on-premise infrastructure

This short getting started description refers to Hopsworks’ Serverless, the free offering. If you would like to use  managed Hopsworks you can contact AI Sweden to access the existing test bed, if you need a specific managed or on-prem deployment , please contact the Hopsworks team or visit hopsworks.ai for more information. 

Getting started:

1. Create an account on app.hopsworks.ai and login. 
    
2. Familiarize yourself with the UI. Learn to navigate to the different capabilities through the menu to the left.
    
3. Familiarize yourself with the APIs by running tutorials - Navigate to the tutorials button at the top right of the UI header. Run one of or tutorials on:

• Predict credit card fraud
• Real-time fraud prediction
• Customer churn prediction

Once you have familiarized yourself and run a couple of tutorials you can go ahead and apply Hopsworks’ capabilities for your own use cases. Below, we breakdown the main usage areas of Hopsworks. 

1. Feature Store
As Hopsworks is primarily a feature store you are able to manage your ingested and external features in feature groups.

The feature store has two main abstractions:

• Feature Groups for the ingestion; where features can be stored in Hopsworks or external (Snowflake or GCP) and are the sets of features that use a common primary key, data freshness (minute data of a user, daily customer purchases…) or can be organized by domain specific logic. The feature groups can be enabled to be online for low latency retrieval.
• Feature Views for retrieval; mapping of the features used by a model, they are not a stored set of features but rather the list of those features and their metadata (version, transformation functions, etc..) that is needed to create training data or retrieve during inference.

2. Model Registry
Once you are satisfied with your features, you can use Hopsworks for your training pipeline. You can register and store your models in Hopsworks Model Registry and compare their performance.  

3. Deployment
As a full end to end ML platform, you can also deploy your models for serving using Hopsworks, using our native KServe integration.

Integration
As an open and modular platform, Hopsworks integrates with all major tools in the Data Science ecosystem and can be deployed on all cloud platforms (GCP, AWS and Azure) as a managed solution. 

Hopsworks is a Python-Centric platform, your existing python code and pipeline can be seamlessly used with Hopsworks and on our enterprise solution you can install your own libraries or any framework you wish and work on a local jupyter environment.