Skip to content

Overview

This chapter describes the essentials of projects, experiments, runs, and their connection with users, artifacts, and models.

Projects

Projects are at the top hierarchy level and group several experiments that belong to it. Projects represent a business problem that shall be solved with a machine learning model that is created through the underlying experiments.

Example

Project: House Price Prediction

Please see the projects page for further details.

Experiments

Experiments belong to exactly one project and group several runs that belong to it. Experiments represent an attempt to solve the business problem with the underlying runs that generate several artifacts as well as models.

Example

Project: House Price Prediction

  • Experiment 1: Linear Regression
  • Experiment 2: Random Forest Regression
  • Experiment 3: Random Forest Regression with HP tuning

Please see the experiments page for further details.

Runs

Runs belong to exactly one project and none, one or many experiments. However, it is good practice to assign a run to at least one experiment. They may have input from previous runs and may generate outputs – named artifacts – such as models. Runs also contain parameters and metrics that are key-value pairs to make them comparable.

Example

Project: House Price Prediction

  • Experiment 1: Linear Regression
    • Run 1: Data Preparation
    • Run 2: Test Train Split
    • Run 3: Linear Regression Training
  • Experiment 2: Random Forest Regression
    • Run 1: Data Preparation
    • Run 2: Test Train Split
    • Run 4: Random Forest Regression Training

Please see the runs page for further details.

Artifacts

Artifacts are generated by runs and may consist of one or a group of files. They have a type that can be freely assigned as well as the artifact name.

Example

Experiment 1: Linear Regression

  • Run 1: Data Preparation
    • Artifact 1: housing.csv
  • Run 2: Test Train Split
    • Artifact 1: Pipeline
  • Run 3: Linear Regression Training
    • Artifact 2: Model

Please see the artifacts page for further details.

Models

Models are a special type of artifact named model and have a special status in ML Aide since they are the heart of the application.

Example

Experiment 1: Linear Regression

  • Run 1: Linear Regression Training 1
    • Artifact 1: Model 1
  • Run 2: Linear Regression Training 2
    • Artifact 1: Model 2

Please see the models page for further details.

Project Settings

Project members can be managed via project settings. The following roles are available:

  • Owner
  • Contributor
  • Viewer

Please see the project settings page for further details.

Users

ML Aide provides internal user management that allows to

  • Update personal information
  • Manage API Keys

Please see the users page for further details.