Cheat Sheet

Basic Step

Connect to workspace

from azureml.core import Workspace
ws = Workspace.from_config()

The workspace object is the primary handle on your Azure ML assets, and it's utilised throughout ws (it's also known as "workspace").

Connect to the Compute Target

compute_target = ws.compute_targets['<compute-target-name>']

Sample Usage

compute_target = ws.compute_targets['powerful-gpu']

config = ScriptRunConfig(
    compute_target=compute_target,  # compute target used to run train.py script
    source_directory='.',
    script='train.py',
)

the complete code for the workload setup

from azureml.core import Workspace
ws = Workspace.create(name='your_workspace', #this will be created and will be added to your system in Azure ML Studio
                      subscription_id='<azure-subscription-id>', #make sure that you have it has a Secret
                      resource_group='your_resourcegroup',
                      create_resource_group=True,
                      location='eastus2' #location do refer the documentation of the Azure
                     )

Prepare Python Environemnt

from azureml.core import Environment
# Option 1. From pip
environment = Environment.from_pip_requirements('<env-name>', '<path/to/requirements.txt - in ml Studio>')
# Option 2. From Conda
environment = Environment.from_conda_specification('<env-name>', '<path/to/env.yml - in ml Studio>')

You can use a pip requirements.txt file or a Conda env.yml file to define a Python environment on your compute.

Sample Code

environment = Environment.from_pip_requirements('<env-name>', '<path/to/requirements.txt - in ml Studio>')

config = ScriptRunConfig(
    environment=environment,  # set the python environment
    source_directory='.',
    script='train.py',
)

you can get to know more about the command line argument

from azureml.core import ScriptRunConfig

config = ScriptRunConfig(
    source_directory='<path/to/code - in ml Studio>',  # relative paths okay
    script='script.py',
    compute_target=compute_target,
    environment=environment,
    arguments=arguments,
)

To run $ (env) python <path/to/code>/script.py [arguments] - in ml Studio> on a remote compute cluster target: ComputeTarget with an environment env: Environment we can use the ScriptRunConfig class

from azureml.core import ScriptRunConfig

config = ScriptRunConfig(
    source_directory='<path/to/code - in ml Studio>',  # relative paths okay
    script='script.py',
    compute_target=compute_target,
    environment=environment,
    arguments=arguments,
)

Commands for the Configuration

compute_target : If not provided the script will run on your local machine/Desktop environment : If not provided, uses a default Python environment managed by Azure ML. See Environment.

command = 'echo cool && python script.py'.split()

config = ScriptRunConfig(
    source_directory='<path/to/code - in ml Studio>',  # relative paths some times will be avaialbe as the Auto Fill
    command=command,
    compute_target=compute_target,
    environment=environment,
    arguments=arguments,
)

Experiments in Azure ML

exp = Experiment(ws, '<experiment-name - in ml Studio>')
run = exp.submit(config)
print(run.get_portal_url())
#In Azure ML Studio where you can monitor your run.

Some Commands to think about

cmd
$ conda env create -f env.yml  # create environment called pytorch
$ conda activate pytorch
(pytorch) $ cd <path/to/code - in ml Studio>
(pytorch) $ python train.py --learning_rate 0.001 --momentum 0.9

Now to run GPU in the Azure ML

ws = Workspace.from_config()
compute_target = ws.compute_targets['powerful-gpu']
environment = Environment.from_conda_specification('pytorch', 'env.yml')

config = ScriptRunConfig(
    source_directory='<path/to/code - in ml Studio>',
    script='train.py',
    environment=environment,
    arguments=['--learning_rate', 0.001, '--momentum', 0.9],
)

run = Experiment(ws, 'PyTorch model training').submit(config)

on Training using the Distributed Training

from azureml.core import Workspace, Experiment, ScriptRunConfig
from azureml.core import Environment
from azureml.core.runconfig import MpiConfiguration

ws = Workspace.from_config()
compute_target = ws.compute_targets['powerful-gpu']
environment = Environment.from_conda_specification('pytorch', 'env.yml')
environment.docker.enabled = True
environment.docker.base_image = 'mcr.microsoft.com/azureml/openmpi3.1.2-cuda10.1-cudnn7-ubuntu18.04'

# train on 2 nodes each with 4 GPUs
mpiconfig = MpiConfiguration(process_count_per_node=4, node_count=2)

config = ScriptRunConfig( 
    source_directory='<path/to/code - in ml Studio>',  # directory containing train.py
    script='train.py',
    environment=environment,
    arguments=['--learning_rate', 0.001, '--momentum', 0.9],
    distributed_job_config=mpiconfig,   # add the distributed configuration
)

run = Experiment(ws, 'PyTorch model training').submit(config)

This will be for the ScriptRunConfig

Here mcr.microsoft.com/azureml/openmpi3.1.2-cuda10.1-cudnn7-ubuntu18.04 : is a docker image with OpenMPI. This is required for distributed training on Azure ML.

MpiConfiguration : is where you specify the number of nodes and GPUs (per node) you want to train on.

Connecting to the Data in Azure ML Studio

datastore = ws.get_default_datastore()
dataset = Dataset.File.from_files(path=(datastore, '<path/on/datastore> - in ml Studio'))

To work with data in your training scripts using your workspace ws and its default datastore:

Now Just pass this to your training script as a command line argument.

arguments=['--data', dataset.as_mount()]