This repo contains ressources that are necessary for AzureML SDK v2.0 Training.
Why this migration to version 2.0 ?
With version 2, it's easier to developp without compute instance.
It can be very interesting because compute instances are sometimes unstable.
It is necessary to have a working WSL environment in order to use this repository.
For that you can :
- Use Michelin Documentation (Michelin Computer)
- Use Microsoft Documentation (Other)
The experience will be better if you have :
- VSCode installed and setup with WSL - Documentation
- Miniconda installed and setup in wsl
Create an environment and install dependancies
conda create -n azml_training python=3.8 openssl=1.1.1n
conda activate azml_training
pip install -r requirements.txtAdd env variable to conda env :
conda env config vars set AZUREML_SUBSCRIPTION='<TENANT-ID>'
conda env config vars set AZUREML_RESSOURCE_GROUP='<RESSOURCE-GROUP>'
conda env config vars set AZUREML_WORKSPACE_NAME='<AML-WORKSPACE>'
conda env config vars listThe aim of all theses env variable is to target the right azureml workspace during your training session.
All theses information are available in your AML workspace url.
In this part we'll talk about how to use Data in AzureML.
In this part, we'll talk about how to create environment in AzureML.
In this part, we'll talk about jobs in AzureML.
In this part, we'll talk about chaining jobs in AzureML Pipelines.