Note
For starting runs from the mantik GUI, you need credentials to one of our supported high-perfomance clusters (HPCs). We support HPCs from Jülich Supercomputing Centre (JSC), Swiss National Supercomputing Centre (CSCS), E4 Computer Engineering.
Without that you can, we will show you how to submit a run on any other infrastructure and using mantik GUI for managing your AI development.
First steps¶
This section takes you through setting up a basic mantik project that can be used for tracking your machine learning experiments. Follow this step-by-step guide to have everything set up for your machine learning journey with mantik.
You will learn the following:
Create a mantik project.
Create an Experiment.
Add a code repository.
Note
No external credentials are needed to run this tutorial.
The UI, currently supports fully mantik on HPC, running projects locally is fully supported through the CLI.
Sign-up with mantik¶
mantik can be used for machine learning experiment tracking only with a mantik account. In case you haven’t signed-up already:
If you have already mantik account, log-in and follow the next steps.
Create a project¶
Once logged in, you are able to create a Project. Follow these steps:
Go straight to the mantik Projects section.
Click on the
+ CREATE PROJECT
buttonFill the form with:
name
short description (optional)
long description (optional)
private/public
Example Inputs
name: My first projectshort description: Tutorials
long description: A tutorial project created when following the step-by-step guide from the mantik official docs.
private (In order to keep the project private, do not click on
Make this a Public Project
)Click on
CREATE
. You will be navigated to the project overview page.
Note
Now that you have created a project it appears in the list of projects accessible to you, displayed in the left column of the Projects page of mantik.
Create an Experiment¶
A Project contains experiments, which are tracked for reproducibility. AnExperiment contains one or several Runs.
To create an Experiment:
Go to your project.
Click on
Experiments
on the side menu.Click on the
+ ADD
button above the table.Provide a name to your Experiment, e.g. my-first-experiment
Click on
CREATE
.
Add code repository to mantik¶
mantik supports bookkeeping the version of the code that you use for your machine learning experimentation. To do so, the code’s git repository has to be added to your mantik project. mantik supports public and private git repositories hosted by GitLab, GitHub and Bitbucket, or on premise.
Here, you will use a demo stored in our public tutorials repo on GitLab: https://gitlab.com/mantik-ai/tutorials, called Wine quality estimator. See also the Kaggle project.
To add a Code repository:
Go to your project.
Click on
Code
on the side menu.Click on the
+ ADD
button (upper-right corner).Fill the form with:
name: mantik Tutorials
description: Public code repository for mantik tutorials.
click on
CREATE