Azure ML Tutorial

Contents
Introduction .................................................................................................................................................. 1
Create your workspace ................................................................................................................................. 2
Create your first Machine Learning Experiment ........................................................................................... 6
Importing data .......................................................................................................................................... 7
Simple Pre-Processing ............................................................................................................................. 12
Actual Machine Learning ........................................................................................................................ 16
Results ..................................................................................................................................................... 20
Publishing your experiment ........................................................................................................................ 21
Creating a Windows 10 application (UWP)................................................................................................. 24
Wrap up ...................................................................................................................................................... 32

Introduction
This tutorial aims to introduce you to AzureML and show you how you can easily create your own
experiment and publish it into a service. We will do this through an example. We are going to use a
dataset containing vote information from US elections at 1984 and make a simple Windows 10
Application “predicting” whether you would be a democrat or a republican at that time.

This tutorial is meant for people that now about Machine Learning and would like to see how this
platform works and what it has to offer but it is also meant for beginners that would like to get started.
However, the purpose of the tutorial is not to teach data analysis but to provide an overview of the tools
and the platform offered by Microsoft Azure.

Links:

Github: https://github.com/georschi/Azure-Machine-Learning-and-UWP-application

Machine Learning Experiment:

http://gallery.cortanaintelligence.com/Experiment/AzureMLTutorialExperiment-1

Georgios Schinas - November 2016

Create your workspace
The first thing you need to do is to create an account at the Azure Machine Learning Workspace.

You can visit https://studio.azureml.net and sign up for free with a Microsoft Account, without a need of
having an Azure Subscription.

Georgios Schinas - November 2016

In case you do have an active Azure Subscription you can create your workspace through the portal as is
shown below

Georgios Schinas - November 2016

Georgios Schinas - November 2016
After a few second the deployment will have successfully finished and you can now head to you ML
Workspace to start our project.

Georgios Schinas - November 2016

Create your first Machine Learning Experiment

Georgios Schinas - November 2016

As soon as you create your new blank experiment you will come across a screen like the one above. At
the top (1) you can see the name of your experiment. By clicking it we will change it into
AzureMLTutorialExperiment. At the middle (2) you may see our canvas. The place where all your
creations will come into life. At the far right (3) is the properties windows. From there we will be
changing some settings to the modules we will include. Finally, at the left (4) there is list of all the things
(datasets, models, tools etc.) that we can include in our experiments.

Importing data
A great benefit of AzureML is the ease at which we can bring data into our experiment. You can either
have your own datasets saved online or import them from anywhere in the internet (you can connect it
to your SQL database if you would like to). We will import our data from a webpage.

Georgios Schinas - November 2016

 To use this module, we drag and drop it into the middle of
our screen (or even double click it).

Now click it and notice the properties window. Change it

so it seems like mine.

The URL we use is https://archive.ics.uci.edu/ml/machine-

learning-databases/voting-records/house-votes-84.data
For your reference you can visit http://archive.ics.uci.edu/ml/ to find a
lot of databases on which you can test your skills

Now it’s time for our first RUN! Notice the bar at the bottom of your screen and press RUN.

After a while the experiment will have run and now by clicking this little circle you can visualize the
results

Georgios Schinas - November 2016

Georgios Schinas - November 2016
There are a few things to notice here.

You can see that we have not names for our columns and there are some missing values. We will deal
with this right now.

Let’s enter manually the names of our columns.

Georgios Schinas - November 2016

In there, we paste the names of the columns that we want to add.

column_name
Class
handicapped-infants
water-project-cost-sharing
adoption-of-the-budget-resolution
physician-fee-freeze
el-salvador-aid
religious-groups-in-schools
anti-satellite-test-ban
aid-to-nicaraguan-contras
mx-missile
immigration
synfuels-corporation-cutback
education-spending
superfund-right-to-sue
crime
duty-free-exports
export-administration-act-south-africa

Georgios Schinas - November 2016

Simple Pre-Processing
For now I have something special for you. You can execute your own R and Python scripts in your
experiment! To do so, we import the module as shown in the picture

Drag and connect the modules as shown in the picture. This way you determine what the input to your
script will be.

Georgios Schinas - November 2016

On the properties window you can write your own code.

The variables dataset1 and dataset2 contain the datasets we gave as input.

The lines 5&6 change the name of the columns and then we delete all those lines that contain a missing
character. NOTE that you could do something else with the missing values, but it’s not the purpose of
this tutorial to deal with these issues.

colnames(dataset2) <- c(dataset1['column_name'])$column_name;

dataset2 = na.omit(dataset2)

By running again the experiment and visualizing the result you can see that our dataset is now ready for
machine learning.

Georgios Schinas - November 2016

Now, let’s use another useful tool, the Filter Based Feature Selection to reduce the features that will
train our ML model. To do so, follow the simple steps:

Georgios Schinas - November 2016

Georgios Schinas - November 2016
Actual Machine Learning
It’s time for real ML now. We need to split our data into training and test sets, train our model and
compare it with our test set.

Let’s do it!

Georgios Schinas - November 2016

Georgios Schinas - November 2016
We need to select a column to tell the model what feature is it trying to predict.

Up to this point we have trained our model. But we need to see how good it is. So we have 2 more
modules to add.

Georgios Schinas - November 2016

Let’s now run our experiment!

Georgios Schinas - November 2016

Results
The experiment should have run in about one minute. Let’s see the results.

Georgios Schinas - November 2016

This beautifully organized page contains a lot of useful information for a data analyst. It provides the
ROC curve, the confusion matrix and on top of that it provides the analyst with a threshold slider where
you can change the threshold of your model and see in real time how the other metrics change!

For our example, we see some great results. One could argue that these results are not representative
since we omitted all those rows with missing values that could substantially contribute to the
performance of our model. This might be the case but again it is out of the scope of this tutorial.

Our model has been trained and therefor now it’s time to publish it and see how to integrate it within an
app.

Publishing your experiment

What we want to do is to setup a web service. Just hover over the SET UP WEB SERVICE and click the
first option. In case you are not able to do that, you might need to re-run your experiment and then you
will be good to go. We click and wait for a bit.

Your main screen should be something like this

Georgios Schinas - November 2016

We need to change where the input of the web service is and place it above the Score Model. (in case
you needed to do some preprocessing with your input you would need to place it an appropriate place)

Georgios Schinas - November 2016

Since we made this change we need to press RUN again. Then just press Deploy Web Service [Classic]

In a moment, you will arrive in a page like the following

Note your API Key since we are going to need it. By pressing Test a window pops up and asks us to enter
values for our input and by clicking the ok button we get a result.

Georgios Schinas - November 2016

At this point if you know how to make http calls from your code you are good to go. You’ve learnt all you
need it to get started with Azure Machine Learning.

Creating a Windows 10 application (UWP)

We start by opening Visual Studio Community and start a New Project.

Georgios Schinas - November 2016

The first thing we should to is to add the nugget package that will help us with the web service calls.

Follow my lead!

Georgios Schinas - November 2016

Let’s go now to the MainPage.xaml

Georgios Schinas - November 2016

What we want to do is to create a layout as simple as possible. So we add to our Grid a few grid rows
and columns, by adding the following lines of code under the Grid. For consistency with my code, give a
name to the grid myGrid. Please go on ahead and copy paste the code either from below or from github.
<Grid.RowDefinitions>
<RowDefinition Height="2*"/>
<RowDefinition/>
<RowDefinition/>
<RowDefinition/>
<RowDefinition/>
<RowDefinition/>
<RowDefinition/>
<RowDefinition Height="2*"/>
</Grid.RowDefinitions>
<Grid.ColumnDefinitions>
<ColumnDefinition Width="*"/>
<ColumnDefinition Width="3*"/>
<ColumnDefinition Width="3*"/>
<ColumnDefinition Width="*"/>
</Grid.ColumnDefinitions>

Below these definitions we will add the following code that in the one row puts the questions and at the
other it give the user the freedom to choose between yes and no.

<TextBlock Grid.ColumnSpan="2" Grid.Column="1" Text="I can predict if you are a

democrat or a republican. Try me!" HorizontalAlignment="Center"
VerticalAlignment="Center" TextWrapping="Wrap"/>
<TextBlock Grid.Column="1" Grid.Row="1" Text="Do you agree with the abortion of
handicapped infants?" TextWrapping="Wrap" VerticalAlignment="Center"/>
<TextBlock Grid.Column="1" Grid.Row="2" Text="Do you agree with water project cost
sharing?" TextWrapping="Wrap" VerticalAlignment="Center"/>
<TextBlock Grid.Column="1" Grid.Row="3" Text="Do you agree with the adoption of the
budget resolution?" TextWrapping="Wrap" VerticalAlignment="Center"/>
<TextBlock Grid.Column="1" Grid.Row="4" Text="Do you agree with the physician fee
freeze?" TextWrapping="Wrap" VerticalAlignment="Center"/>
<TextBlock Grid.Column="1" Grid.Row="5" Text="Do you agree with the El-Salvador aid?"
TextWrapping="Wrap" VerticalAlignment="Center"/>
<TextBlock Grid.Column="1" Grid.Row="6" Text="Do you agree with the religious groups in
schools?" TextWrapping="Wrap"/>

<ToggleSwitch Grid.Column="2" Grid.Row="1" OffContent="NO" OnContent="YES"

HorizontalAlignment="Right" VerticalAlignment="Center" />
<ToggleSwitch Grid.Column="2" Grid.Row="2" OffContent="NO" OnContent="YES"
HorizontalAlignment="Right" VerticalAlignment="Center"/>
<ToggleSwitch Grid.Column="2" Grid.Row="3" OffContent="NO" OnContent="YES"
HorizontalAlignment="Right" VerticalAlignment="Center" />
<ToggleSwitch Grid.Column="2" Grid.Row="4" OffContent="NO" OnContent="YES"
HorizontalAlignment="Right" VerticalAlignment="Center"/>
<ToggleSwitch Grid.Column="2" Grid.Row="5" OffContent="NO" OnContent="YES"
HorizontalAlignment="Right" VerticalAlignment="Center"/>
<ToggleSwitch Grid.Column="2" Grid.Row="6" OffContent="NO" OnContent="YES"
HorizontalAlignment="Right" VerticalAlignment="Center"/>
<Button Grid.Column="1" Grid.ColumnSpan="2" Grid.Row="7" HorizontalAlignment="Center"
Content="Submit" Click="Button_Click" />

Georgios Schinas - November 2016

Your screen should look like this

Now, open the MainPage.xaml.cs and right before the constructor of the MainPage declare a variable
private string[] answers = new string[6];

Georgios Schinas - November 2016

Now return to your browser and click REQUEST/RESPONSE as shown below, which will take to a very
helpful page.

That page explains exactly what happens with the http calls and gives you hints and help to implement
them in any platform, so if interested, give it some time. For our project we will take advantage of some
C# code that has been regenerated for us. Simply scroll down a bit and you’ll find it.

From that code, copy the function static async Task InvokeRequestResponseService()
which contains the basic structure to call the service. After you copy-paste it there will be a few error
indications but don’t worry! That’s only because this C# code was generated for a console application
and not a UWP. Make a few changes to your code so that it looks like the one below (or you directly that
instead)
private async Task InvokeRequestResponseService()
{
using (var client = new HttpClient())
{
var scoreRequest = new
{
Inputs = new Dictionary<string, StringTable>() {

Georgios Schinas - November 2016

 {
"input1",
new StringTable()
{
ColumnNames = new string[] {"Class", "handicapped-
infants", "water-project-cost-sharing", "adoption-of-the-budget-resolution",
"physician-fee-freeze", "el-salvador-aid", "religious-groups-in-schools"},
Values = new string[,] { { "null", answers[0],
answers[1], answers[2], answers[3], answers[4], answers[5] }, }
}
},
},
GlobalParameters = new Dictionary<string, string>()
{
}
};
const string apiKey = ""; // Replace this with your API key
client.DefaultRequestHeaders.Authorization = new
AuthenticationHeaderValue("Bearer", apiKey);

client.BaseAddress = new Uri("enter your own uri here as in the sample

code");

HttpResponseMessage response = await client.PostAsJsonAsync("",

scoreRequest);

if (response.IsSuccessStatusCode)
{
string json = await response.Content.ReadAsStringAsync();
dynamic result = JsonConvert.DeserializeObject<dynamic>(json);
var whatAreYou = result.Results.output1.value.Values[0][7];
var thatMuch = result.Results.output1.value.Values[0][8];
float percent = (float)thatMuch;
percent = (percent < 0.5) ? (1 - percent) * 100 : percent * 100;
percent = (int)percent;
var dialog = new MessageDialog(String.Format("You are a {0} and I'm
{1}% sure about it!", whatAreYou, percent));
await dialog.ShowAsync();

}
else
{
// custom code to deal with a problem at the call
}
}
}

In that code what you need to notice to put your own API Key and Uri. The errors that remain will go
away once you add this class in your code
public class StringTable
{
public string[] ColumnNames { get; set; }
public string[,] Values { get; set; }
}

Georgios Schinas - November 2016

and after you add these lines at the top of the page
using System.Threading.Tasks;
using System.Net.Http;
using System.Net.Http.Headers;
using Newtonsoft.Json;
using Windows.UI.Popups;

There is one final thing to add here and we are done! This is nothing other than calling that function and
filling up the variable answers[] with the appropriate content, when the button submit is clicked.

To do so, add the following code and you’re good to go (when writing the xaml part you may have
noticed that at the button we already added the click event Click="Button_Click").

private async void Button_Click(object sender, RoutedEventArgs e)

{
foreach (FrameworkElement item in myGrid.Children)
{
if (item is ToggleSwitch) answers[Grid.GetRow(item) - 1] =
(((ToggleSwitch)item).IsOn == true) ? "y" : "n";
}
await InvokeRequestResponseService();
}

What this function does is to search all the ToggleSwitch items we have and put their respective value in
the variable answers[] at the correct position. You may need to take some time to agree with that line of
code, don’t worry!

Run the application and have fun, it should look like this:

Georgios Schinas - November 2016

Note: my answers here where chosen at random, nothing personal :P

Wrap up
If you’ve reached this point you have successfully finished this tutorial. Congratulations!

Thanks for staying with me throughout my first tutorial. Please feel free to leave any comments or
questions, See you soon!

Georgios Schinas - November 2016