Azure Machine Learning (AML) Examples

Welcome to the Azure Machine Learning (AML) examples repository!

Prerequisites

1. An Azure subscription. If you don't have an Azure subscription, create a free account before you begin.
2. Familiarity with Python and Azure Machine Learning concepts.
3. A terminal and Python >=3.6,<3.9.

Installation

Clone this repository and install required packages:

git clone https://github.com/Azure/azureml-examples
cd azureml-examples
pip install --upgrade -r requirements.txt

To create or setup a workspace with the assets used in these examples, run the setup script.

If you do not have an AML Workspace, run python setup-workspace.py --subscription-id $ID, where $ID is your Azure subscription id. A resource group, AML Workspace, and other necessary resources will be created in the subscription.

If you have an AML Workspace, install the AML CLI and run az ml folder attach -w $WS -g $RG, where $WS and $RG are the workspace and resource group names.

Run python setup-workspace.py -h to see other arguments.

Contents

This example repo is structured for real ML projects, with modifications specific to showing examples. You can use the official GitHub template to schedule your own ML workflow(s).

directory	description
.cloud	cloud templates
.github	GitHub specific files like Actions workflow yaml definitions and issue templates
code	ML code organized by scenario (train, deploy, etc.) then tool (pytorch, tensorflow, etc.)
data	not recommended - used for convenient data for examples - data should not be stored directly in a repository
environments	environment definition files such as conda yaml, pip txt, or dockerfile
mlprojects	mlflow project specifications
models	not recommended - used for convenient models for examples - models should not be stored directly in a repository
notebooks	interactive jupyter notebooks for iterative ML development
tutorials	not recommended - end to end tutorials
website	not recommended - used for hosting website
workflows	AML control plane specification (currently Python scripts) of job(s) to be run

Getting started

To get started, try the introductory tutorial.

Contributing

We welcome contributions and suggestions! Please see the contributing guidelines for details.

Code of Conduct

This project has adopted the Microsoft Open Source Code of Conduct. Please see the code of conduct for details.

Index of Examples

Tutorials

path	status	notebooks	description
an-introduction		1.hello-world.ipynb 2.pytorch-model.ipynb 3.pytorch-model-cloud-data.ipynb	learn the basics of Azure Machine Learning
automl-with-pycaret		1.classification.ipynb	learn how to automate ML with pycaret
deploy-edge		ase-gpu.ipynb	learn how to use Edge device for model deployment and scoring
deploy-triton		1.densenet-local.ipynb 2.bidaf-aks-v100.ipynb	learn how to efficiently deploy to GPUs using triton inference server
music-with-ml		1.intro-to-magenta.ipynb	learn how to create music with ML using magenta
using-dask		1.intro-to-dask.ipynb 2.eds-at-scale.ipynb	learn how to read from cloud data and scale PyData tools (numpy, pandas, scikit-learn, etc.) with dask
using-mlflow		sklearn.ipynb	learn how to use AML as the backend for mlflow
using-optuna		1.intro-to-optuna.ipynb	learn how to optimize an objective function with optuna
using-pytorch-lightning		1.train-single-node.ipynb 2.log-with-tensorboard.ipynb 3.log-with-mlflow.ipynb 4.train-multi-node-ddp.ipynb	learn how to train and log metrics with PyTorch Lightning
using-rapids		1.train-and-hpo.ipynb 2.train-multi-gpu.ipynb	learn how to accelerate PyData tools (numpy, pandas, scikit-learn, etc) on NVIDIA GPUs with rapids

Notebooks

path	description
notebooks/train-lightgbm-local.ipynb	use AML and mlflow to track interactive experimentation in the cloud

Train

path	compute	environment	description
workflows/train/deepspeed-cifar.py	AML - GPU	docker	train CIFAR-10 using DeepSpeed and PyTorch
workflows/train/fastai-mnist-mlproject.py	AML - CPU	mlproject	train fastai resnet18 model on mnist data via mlflow mlproject
workflows/train/fastai-mnist.py	AML - CPU	conda	train fastai resnet18 model on mnist data
workflows/train/fastai-pets.py	AML - GPU	docker	train fastai resnet34 model on pets data
workflows/train/lightgbm-iris.py	AML - CPU	pip	train a lightgbm model on iris data
workflows/train/pytorch-mnist-mlproject.py	AML - GPU	mlproject	train a pytorch CNN model on mnist data via mlflow mlproject
workflows/train/pytorch-mnist.py	AML - GPU	conda	train a pytorch CNN model on mnist data
workflows/train/sklearn-diabetes-mlproject.py	AML - CPU	mlproject	train sklearn ridge model on diabetes data via mlflow mlproject
workflows/train/sklearn-diabetes.py	AML - CPU	conda	train sklearn ridge model on diabetes data
workflows/train/tensorflow-iris.py	AML - CPU	conda	train tensorflow NN model on iris data
workflows/train/tensorflow-mnist-distributed-horovod.py	AML - GPU	conda	train tensorflow CNN model on mnist data distributed via horovod
workflows/train/tensorflow-mnist-distributed.py	AML - GPU	conda	train tensorflow CNN model on mnist data distributed via tensorflow
workflows/train/tensorflow-mnist.py	AML - GPU	conda	train tensorflow NN model on mnist data
workflows/train/xgboost-iris.py	AML - CPU	pip	train xgboost model on iris data

Deploy

path	compute	description
workflows/deploy/pytorch-mnist-aks-cpu.py	AKS - CPU	deploy pytorch CNN model trained on mnist data to AKS
workflows/deploy/sklearn-diabetes-aks-cpu.py	AKS - CPU	deploy sklearn ridge model trained on diabetes data to AKS

Reference

• Azure Machine Learning Documentation
• Python SDK Documentation
• Azure Machine Learning Pipelines Notebooks