Tutorial: Train an image classification model with Azure Machine Learning service

In this tutorial, you train a machine learning model both locally and on remote compute resources. You use the training and deployment workflow for Azure Machine Learning service in a Python Jupyter notebook. You can then use the notebook as a template to train your own machine learning model with your own data. This tutorial is part one of a two-part tutorial series.

This tutorial trains a simple logistic regression by using the MNIST dataset and scikit-learn with Azure Machine Learning service. MNIST is a popular dataset consisting of 70,000 grayscale images. Each image is a handwritten digit of 28 x 28 pixels, representing a number from zero to nine. The goal is to create a multiclass classifier to identify the digit a given image represents.

Learn how to take the following actions:

  • Set up your development environment.
  • Access and examine the data.
  • Train a simple logistic regression locally by using the popular scikit-learn machine learning library.
  • Train multiple models on a remote cluster.
  • Review training results and register the best model.

You learn how to select a model and deploy it in part two of this tutorial.

If you don’t have an Azure subscription, create a free account before you begin. Try the free or paid version of Azure Machine Learning service today.


Code in this article was tested with Azure Machine Learning SDK version 1.0.2.

Get the notebook

For your convenience, this tutorial is available as a Jupyter notebook. Run the tutorials/img-classification-part1-training.ipynb notebook either in Azure Notebooks or in your own Jupyter notebook server.

Learn how to run notebooks by following the article, Use Jupyter notebooks to explore Azure Machine Learning service.

Set up your development environment

All the setup for your development work can be accomplished in a Python notebook. Setup includes the following actions:

  • Import Python packages.
  • Connect to a workspace, so that your local computer can communicate with remote resources.
  • Create an experiment to track all your runs.
  • Create a remote compute target to use for training.

Import packages

Import Python packages you need in this session. Also display the Azure Machine Learning SDK version:

%matplotlib inline
import numpy as np
import matplotlib
import matplotlib.pyplot as plt

import azureml
from azureml.core import Workspace, Run

# check core SDK version number
print("Azure ML SDK Version: ", azureml.core.VERSION)

Connect to a workspace

Create a workspace object from the existing workspace. Workspace.from_config() reads the file config.json and loads the details into an object named ws:

# load workspace configuration from the config.json file in the current folder.
ws = Workspace.from_config()
print(ws.name, ws.location, ws.resource_group, ws.location, sep = '\t')

Create an experiment

Create an experiment to track the runs in your workspace. A workspace can have multiple experiments:

experiment_name = 'sklearn-mnist'

from azureml.core import Experiment
exp = Experiment(workspace=ws, name=experiment_name)

Create or attach an existing AMlCompute

By using Azure Machine Learning Compute (AmlCompute), a managed service, data scientists can train machine learning models on clusters of Azure virtual machines. Examples include VMs with GPU support. In this tutorial, you create AmlCompute as your training environment. This code creates the compute clusters for you if they don't already exist in your workspace.

Creation of the compute takes about five minutes. If the compute is already in the workspace, this code uses it and skips the creation process:

from azureml.core.compute import AmlCompute
from azureml.core.compute import ComputeTarget
import os

# choose a name for your cluster
compute_name = os.environ.get("AML_COMPUTE_CLUSTER_NAME", "cpucluster")
compute_min_nodes = os.environ.get("AML_COMPUTE_CLUSTER_MIN_NODES", 0)
compute_max_nodes = os.environ.get("AML_COMPUTE_CLUSTER_MAX_NODES", 4)

# This example uses CPU VM. For using GPU VM, set SKU to STANDARD_NC6
vm_size = os.environ.get("AML_COMPUTE_CLUSTER_SKU", "STANDARD_D2_V2")

if compute_name in ws.compute_targets:
    compute_target = ws.compute_targets[compute_name]
    if compute_target and type(compute_target) is AmlCompute:
        print('found compute target. just use it. ' + compute_name)
    print('creating a new compute target...')
    provisioning_config = AmlCompute.provisioning_configuration(vm_size = vm_size,
                                                                min_nodes = compute_min_nodes, 
                                                                max_nodes = compute_max_nodes)

    # create the cluster
    compute_target = ComputeTarget.create(ws, compute_name, provisioning_config)

    # can poll for a minimum number of nodes and for a specific timeout. 
    # if no min node count is provided it will use the scale settings for the cluster
    compute_target.wait_for_completion(show_output=True, min_node_count=None, timeout_in_minutes=20)

     # For a more detailed view of current AmlCompute status, use the 'status' property    

You now have the necessary packages and compute resources to train a model in the cloud.

Explore data

Before you train a model, you need to understand the data that you use to train it. You also need to copy the data into the cloud. Then it can be accessed by your cloud training environment. In this section, you learn how to take the following actions:

  • Download the MNIST dataset.
  • Display some sample images.
  • Upload data to the cloud.

Download the MNIST dataset

Download the MNIST dataset and save the files into a data directory locally. Images and labels for both training and testing are downloaded:

import os
import urllib.request

os.makedirs('./data', exist_ok = True)

urllib.request.urlretrieve('http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz', filename='./data/train-images.gz')
urllib.request.urlretrieve('http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz', filename='./data/train-labels.gz')
urllib.request.urlretrieve('http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz', filename='./data/test-images.gz')
urllib.request.urlretrieve('http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz', filename='./data/test-labels.gz')

Display some sample images

Load the compressed files into numpy arrays. Then use matplotlib to plot 30 random images from the dataset with their labels above them. This step requires a load_data function that's included in an util.py file. This file is included in the sample folder. Make sure it's placed in the same folder as this notebook. The load_data function simply parses the compressed files into numpy arrays:

# make sure utils.py is in the same directory as this code
from utils import load_data

# note we also shrink the intensity values (X) from 0-255 to 0-1. This helps the model converge faster.
X_train = load_data('./data/train-images.gz', False) / 255.0
y_train = load_data('./data/train-labels.gz', True).reshape(-1)

X_test = load_data('./data/test-images.gz', False) / 255.0
y_test = load_data('./data/test-labels.gz', True).reshape(-1)

# now let's show some randomly chosen images from the training set.
count = 0
sample_size = 30
plt.figure(figsize = (16, 6))
for i in np.random.permutation(X_train.shape[0])[:sample_size]:
    count = count + 1
    plt.subplot(1, sample_size, count)
    plt.text(x=10, y=-10, s=y_train[i], fontsize=18)
    plt.imshow(X_train[i].reshape(28, 28), cmap=plt.cm.Greys)

A random sample of images displays:

Random sample of images

Now you have an idea of what these images look like and the expected prediction outcome.

Upload data to the cloud

Now make the data accessible remotely by uploading that data from your local machine into Azure. Then it can be accessed for remote training. The datastore is a convenient construct associated with your workspace for you to upload or download data. You can also interact with it from your remote compute targets. It's backed by an Azure Blob storage account.

The MNIST files are uploaded into a directory named mnist at the root of the datastore:

ds = ws.get_default_datastore()
print(ds.datastore_type, ds.account_name, ds.container_name)

ds.upload(src_dir='./data', target_path='mnist', overwrite=True, show_progress=True)

You now have everything you need to start training a model.

Train a local model

Train a simple logistic regression model by using scikit-learn locally.

Training locally can take a minute or two depending on your computer configuration:

from sklearn.linear_model import LogisticRegression

clf = LogisticRegression()
clf.fit(X_train, y_train)

Next make predictions by using the test set and calculate the accuracy:

y_hat = clf.predict(X_test)
print(np.average(y_hat == y_test))

The local model accuracy displays:


With just a few lines of code, you have 92 percent accuracy.

Train on a remote cluster

Now you can expand on this simple model by building a model with a different regularization rate. This time, you train the model on a remote resource.

For this task, submit the job to the remote training cluster you set up earlier. To submit a job, you take the following steps:

  • Create a directory.
  • Create a training script.
  • Create an estimator object.
  • Submit the job.

Create a directory

Create a directory to deliver the necessary code from your computer to the remote resource:

import os
script_folder = './sklearn-mnist'
os.makedirs(script_folder, exist_ok=True)

Create a training script

To submit the job to the cluster, first create a training script. Run the following code to create the training script called train.py in the directory you created. This training adds a regularization rate to the training algorithm. So it produces a slightly different model than the local version:

%%writefile $script_folder/train.py

import argparse
import os
import numpy as np

from sklearn.linear_model import LogisticRegression
from sklearn.externals import joblib

from azureml.core import Run
from utils import load_data

# let user feed in 2 parameters, the location of the data files (from datastore), and the regularization rate of the logistic regression model
parser = argparse.ArgumentParser()
parser.add_argument('--data-folder', type=str, dest='data_folder', help='data folder mounting point')
parser.add_argument('--regularization', type=float, dest='reg', default=0.01, help='regularization rate')
args = parser.parse_args()

data_folder = os.path.join(args.data_folder, 'mnist')
print('Data folder:', data_folder)

# load train and test set into numpy arrays
# note we scale the pixel intensity values to 0-1 (by dividing it with 255.0) so the model can converge faster.
X_train = load_data(os.path.join(data_folder, 'train-images.gz'), False) / 255.0
X_test = load_data(os.path.join(data_folder, 'test-images.gz'), False) / 255.0
y_train = load_data(os.path.join(data_folder, 'train-labels.gz'), True).reshape(-1)
y_test = load_data(os.path.join(data_folder, 'test-labels.gz'), True).reshape(-1)
print(X_train.shape, y_train.shape, X_test.shape, y_test.shape, sep = '\n')

# get hold of the current run
run = Run.get_context()

print('Train a logistic regression model with regularization rate of', args.reg)
clf = LogisticRegression(C=1.0/args.reg, random_state=42)
clf.fit(X_train, y_train)

print('Predict the test set')
y_hat = clf.predict(X_test)

# calculate accuracy on the prediction
acc = np.average(y_hat == y_test)
print('Accuracy is', acc)

run.log('regularization rate', np.float(args.reg))
run.log('accuracy', np.float(acc))

os.makedirs('outputs', exist_ok=True)
# note file saved in the outputs folder is automatically uploaded into experiment record
joblib.dump(value=clf, filename='outputs/sklearn_mnist_model.pkl')

Notice how the script gets data and saves models:

  • The training script reads an argument to find the directory that contains the data. When you submit the job later, you point to the datastore for this argument: parser.add_argument('--data-folder', type=str, dest='data_folder', help='data directory mounting point').

  • The training script saves your model into a directory named outputs:
    joblib.dump(value=clf, filename='outputs/sklearn_mnist_model.pkl').
    Anything written in this directory is automatically uploaded into your workspace. You access your model from this directory later in the tutorial. The file utils.py is referenced from the training script to load the dataset correctly. Copy this script into the script folder, so that it can be accessed along with the training script on the remote resource.

import shutil
shutil.copy('utils.py', script_folder)

Create an estimator

An estimator object is used to submit the run. Create your estimator by running the following code to define these items:

  • The name of the estimator object, est.
  • The directory that contains your scripts. All the files in this directory are uploaded into the cluster nodes for execution.
  • The compute target. In this case, you use the Azure Machine Learning compute cluster you created.
  • The training script name, train.py.
  • Parameters required from the training script.
  • Python packages needed for training.

In this tutorial, this target is AmlCompute. All files in the script folder are uploaded into the cluster nodes for run. The data_folder is set to use the datastore, ds.as_mount():

from azureml.train.estimator import Estimator

script_params = {
    '--data-folder': ds.as_mount(),
    '--regularization': 0.8

est = Estimator(source_directory=script_folder,

Submit the job to the cluster

Run the experiment by submitting the estimator object:

run = exp.submit(config=est)

Because the call is asynchronous, it returns a Preparing or Running state as soon as the job is started.

Monitor a remote run

In total, the first run takes about 10 minutes. But for subsequent runs, as long as the script dependencies don't change, the same image is reused. So the container startup time is much faster.

What happens while you wait:

  • Image creation: A Docker image is created that matches the Python environment specified by the estimator. The image is uploaded to the workspace. Image creation and uploading takes about five minutes.

    This stage happens once for each Python environment because the container is cached for subsequent runs. During image creation, logs are streamed to the run history. You can monitor the image creation progress by using these logs.

  • Scaling: If the remote cluster requires more nodes to do the run than currently available, additional nodes are added automatically. Scaling typically takes about five minutes.

  • Running: In this stage, the necessary scripts and files are sent to the compute target. Then datastores are mounted or copied. And then the entry_script is run. While the job is running, stdout and the ./logs directory are streamed to the run history. You can monitor the run's progress by using these logs.

  • Post-processing: The ./outputs directory of the run is copied over to the run history in your workspace, so you can access these results.

You can check the progress of a running job in several ways. This tutorial uses a Jupyter widget and a wait_for_completion method.

Jupyter widget

Watch the progress of the run with a Jupyter widget. Like the run submission, the widget is asynchronous and provides live updates every 10 to 15 seconds until the job finishes:

from azureml.widgets import RunDetails

This still snapshot is the widget shown at the end of training:

Notebook widget

Get log results upon completion

Model training and monitoring happen in the background. Wait until the model has finished training before you run more code. Use wait_for_completion to show when the model training is finished:

run.wait_for_completion(show_output=False) # specify True for a verbose log

Display run results

You now have a model trained on a remote cluster. Retrieve the accuracy of the model:


The output shows the remote model has accuracy slightly higher than the local model because of the addition of the regularization rate during training:

{'regularization rate': 0.8, 'accuracy': 0.9204}

In the next tutorial, you explore this model in more detail.

Register model

The last step in the training script wrote the file outputs/sklearn_mnist_model.pkl in a directory named outputs in the VM of the cluster where the job is run. outputs is a special directory in that all content in this directory is automatically uploaded to your workspace. This content appears in the run record in the experiment under your workspace. So the model file is now also available in your workspace.

You can see files associated with that run:


Register the model in the workspace, so that you or other collaborators can later query, examine, and deploy this model:

# register model 
model = run.register_model(model_name='sklearn_mnist', model_path='outputs/sklearn_mnist_model.pkl')
print(model.name, model.id, model.version, sep = '\t')

Clean up resources


The resources you created can be used as prerequisites to other Azure Machine Learning service tutorials and how-to articles.

If you don't plan to use the resources you created, delete them, so you don't incur any charges:

  1. In the Azure portal, select Resource groups on the far left.

    Delete in the Azure portal

  2. From the list, select the resource group you created.

  3. Select Delete resource group.

  4. Enter the resource group name. Then select Delete.

You can also delete just the Azure Machine Learning Compute cluster. However, autoscale is turned on, and the cluster minimum is zero. So this particular resource won't incur additional compute charges when not in use:

# optionally, delete the Azure Machine Learning Compute cluster

Next steps

In this Azure Machine Learning service tutorial, you used Python for the following tasks:

  • Set up your development environment.
  • Access and examine the data.
  • Train a simple logistic regression locally by using the popular scikit-learn machine learning library.
  • Train multiple models on a remote cluster.
  • Review training details and register the best model.

You're ready to deploy this registered model by using the instructions in the next part of the tutorial series: