Azure Machine Learning Training SDK

1.62.0 verified Fri May 01 auth: no python maintenance

The Azure Machine Learning Python SDK training package provides functionality for configuring, submitting, and managing machine learning training runs on Azure. It includes support for hyperparameter tuning (HyperDrive), automated ML (AutoML), and distributed training. Current version is 1.62.0, last updated in early 2025. Release cadence is monthly, aligned with the Azure ML SDK releases.

pip install azureml-train

Common errors

error ModuleNotFoundError: No module named 'azureml.train' ↓

cause Missing azureml-train package or import path mismatch.

fix

Run: pip install azureml-train. Ensure import uses correct submodule, e.g., 'from azureml.train.hyperdrive import ...'.

error ImportError: cannot import name 'ScriptRunConfig' from 'azureml.train' ↓

cause ScriptRunConfig was moved to azureml.core in recent SDK versions.

fix

Change import to: from azureml.core import ScriptRunConfig

error AttributeError: module 'azureml.train.hyperdrive' has no attribute 'HyperDriveRunConfig' ↓

cause HyperDriveRunConfig was renamed to HyperDriveConfig.

fix

Use 'HyperDriveConfig' instead. Import: from azureml.train.hyperdrive import HyperDriveConfig

Warnings

breaking Azure ML SDK v2 is now preferred; v1 (azureml-train) is in maintenance mode. New features only for v2. ↓

fix Migrate to azure-ai-ml (v2). See Microsoft migration guide.

deprecated AutoMLConfig from azureml-train is deprecated in favor of azure-ai-ml automl. ↓

fix Use azure-ai-ml to create automated ML jobs.

gotcha ScriptRunConfig import changed from azureml.train to azureml.core in SDK v1.0.85. ↓

fix Always import ScriptRunConfig from azureml.core.

gotcha HyperDriveConfig replaces deprecated HyperDriveRunConfig. Using old name causes ImportError. ↓

fix Use from azureml.train.hyperdrive import HyperDriveConfig.

Imports

HyperDriveConfig

wrong

from azureml.train.hyperdrive import HyperDriveRunConfig

correct

from azureml.train.hyperdrive import HyperDriveConfig

HyperDriveRunConfig was removed in earlier SDK versions. Use HyperDriveConfig.

AutoMLConfig

wrong

from azureml.train.automl import AutoMLConfig (not from azureml.train or from azureml.core)

correct

from azureml.train.automl import AutoMLConfig

Common mistake: importing from wrong submodule.

ScriptRunConfig

wrong

from azureml.train import ScriptRunConfig

correct

from azureml.core import ScriptRunConfig

ScriptRunConfig moved to azureml.core in SDK v1.0.85+.

Quickstart

Basic example: connect to workspace, create compute target, submit a HyperDrive sweep.

from azureml.core import Workspace, Experiment, ScriptRunConfig
from azureml.core.compute import ComputeTarget, AmlCompute
from azureml.core.compute_target import ComputeTargetException
from azureml.train.hyperdrive import HyperDriveConfig, PrimaryMetricGoal, choice
# Connect to workspace
ws = Workspace.from_config()
# Create a compute cluster (if not exists)
cluster_name = "cpu-cluster"
try:
    compute_target = ComputeTarget(workspace=ws, name=cluster_name)
except ComputeTargetException:
    compute_config = AmlCompute.provisioning_configuration(vm_size='STANDARD_D2_V2', max_nodes=4)
    compute_target = ComputeTarget.create(ws, cluster_name, compute_config)
compute_target.wait_for_completion(show_output=True)
# Configure a hyperparameter sweep
exp = Experiment(ws, 'my-experiment')
src = ScriptRunConfig(source_directory='.', script='train.py', compute_target=compute_target)
hyperdrive_config = HyperDriveConfig(run_config=src,
                                     hyperparameter_sampling=choice({'--lr': [0.01, 0.001]}),
                                     primary_metric_name='accuracy',
                                     primary_metric_goal=PrimaryMetricGoal.MAXIMIZE,
                                     max_total_runs=4)
submitted_run = exp.submit(hyperdrive_config)
submitted_run.wait_for_completion(show_output=True)