Core Concepts

Project

A project is a collection of all artificats in developing ML models. A project consists of folder, annotation labs, pipelines, tracks, models, WebApps, templates, and images.

VC Project

A VC project is a type of project that incorporates version control functions for source code and data.

Folder

A folder (also shown as Data in MLSteam) is a collection of data organized in files and directories. Files in folder, like dateset, could be used in labs for model training and validation.

Workspace

A workspace is a special kind of folder accessible in VC projects that is version controlled and is mounted automatically in VC labs.

Annotation

Annotation provides mainstreaming annotation tools and well integrated to MLSteam..

Lab

A lab is a Web IDE (based on JupyterLab with MLSteam’s add-on functionalities) that organizes files and datasets. You may design ML models and make experiments in a lab. When the development is done, you may convert a lab into a template for reuse in other labs, pipelines or deployment.

Pipeline

A pipeline is a repeatable procedure consisting of actions for running ML tasks. You may define a pipeline for a subset of common ML tasks. You may even define an end-to-end pipeline to fulfill MLOps that retrains and evaluates the model for new model designs or dataset and finally deploys the ML application to an experimental or production site.

Track

A track keeps various results of ML training or experiments, including the parameters, metrics, console logs, and any logged files or data. It also enables visualization of the results with TensorBoard.

Model

A model is a collection of files that record a trained ML model.

WebApp

A WebApp enables deployment of a Web-based ML applications in a simple way. Services for project users may also be provided as a WebApp.

Template

A template is a creator of a lab, pipeline action, or WebApp with predefined programs, datasets, models, or other settings.

Image

An image (Docker image) is used to create a template or to run a container. In MLSteam, an image could be obtained from a user uploaded Docker image file, a remote registry, or an MLSteam-managed registry.

Flavor

A flavor describes how many hardware resources (such as CPUs, GPUs, and memory) are to be allocated.