All About three persons – I ME and MYSELF
All About three persons – I ME and MYSELF

Simple steps towards Azure ML

I have been working with ML in google colab because of the ease and convenience it provides. But it was high time I understood the implementation in a cloud ML offering from one of the major cloud platforms. I was actually fed up with multiple iterations of the same model to reach a respectable accuracy. I decided to chose Azure ML as my first goal and also decided to document my journey. If you intend to follow the same path – HOP ON.

All Azure offerings are grouped as different category groups in the Azure portal. Get an account and reach If it is your first time you would get some credit and a free account.

WORKSPACE. Let us first create a Machine Learning workspace in Azure Portal. Imagine the workspace as the environment boundary I am defining for my ML exercise or experiment. Everything I work on is inside this workspace unless I have explicitly connected to a service/resource from outside.

Chose the subscription to be used.

Mention the resource group. A RESOURCE GROUP is the group I have defined for all the resources I would be using in the workspace. Just as the name says. Neat huh 🙂 . If you don’t have a resource group already, give a name and create a new one.

Give a name to the workspace and chose a region where the workspace would be created. The documentation says to use the region nearest to you, but I would say to go with any US based region because they allow the maximum types of resources. We can later work on closest region approach when we deal with availability and such issues.

Check the other tabs if you want to explore else just leave the defaults and click on Review+Create.

Validation passed – Means you are ready to create your first resource in Azure

Make it a habit to review the names in the review page. Once satisfied go for create.

You can watch the status on the top right
Ort click on the bell icon to check the progress of all activity

Once completed you will see the completion notification. It is pretty simple so I would say go ahead with it right away.

All done. I have my workspace created.

Now let us go to the resource (workspace) and have a look around. Don’t worry about the multiple options in the left-bar. Instead of digging around all the tabs I would suggest to come here as you feel the need for more information or control. They include all the stuff you would ever want to know about your workspace aka silo. Just remember to come here whenever you get anxious about anything happening in the workspace. Or just to know more about everything in your workspace.

With the bed set, lets us go to the ML Studio and indulge in some foreplay.

You can launch the ML Studio to start the ML journey

For names, I would suggest to use a neat mention which is easy for you to connect/recollect later. My naming convention in <context>-<resource-prefix><numeric>. For example my names related to my learning as part of dp100 goes like – dp100-rg01, dp100-wp01 etc

Yes, you noticed it right , I use wp instead of ws for workspace due to my years of association with wordpress(wp). You should use what works for you. My point here is; do have a naming convention which you follow across the environment. You will thank yourself later when you have scores of assets all around.

Next We explore ML Studio ….

When you click to launch the studio you might reach a “workspace not found” page. Select the current subscription and than the workspace we just created and the studio welcome page would come up fine.

The Azure ML Studio landing page provides a glimpse of the four phases of a ML deployment lifecycle namely :

  1. Register data
    Reference data from storage to easily access during model training and explore using summary statistics.
  2. Train models
    Use machine learning algorithms with training data to create models.
  3. Evaluate models
    Find the best model using test data.
  4. Deploy models
    Deploy model as a web service in the Azure cloud, or to IoT Edge devices.

The studio has authoring tools like Notebooks, Automated ML and Designer which allow to build, train, evaluate, and deploy machine learning models.

Assets like Datasets, Experiments, Pipelines, Models and Endpoints are available to prepare the data , run and monitor the jobs, view and deploy the models and manage real-time or batch pipeline endpoints.

The studio also allows us to create, view and manage the compute instances, Datastores and the Data Labeling projects in the workspace.

If we have multiple workspaces we can jump across them using the “Directory + Subscription + Workspace” picker on the top right.

Leave a comment

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.