Azure Machine Learning Designer

/

Azure Machine Learning designer is a visual-first environment that lets you build, test, and deploy, predictive models via a drag and drop interface without needing to write a single line of code. You may recall a similar service announced back in June 2014 called ML Studio, now known as Azure Machine Learning Studio (classic). Designer can be thought of as ML Studio’s successor.

Content

  1. History

  2. User Interface

  3. Concepts

  4. Demonstration

  5. Pipeline Workflows

  6. Compute Targets

  7. Resources

1. History

aml.gif

2. Designer User Interface

  1. Navigation Bar

    e.g. mlworkspace > Designer > Authoring

  2. Module Palette

    Each module represents a set of code that can run independently and perform a machine learning task, given the required inputs. To understand the technical background behind each modules, see http://aka.ms/aml/designer-modules.

  3. Pipeline Settings

    Clicking this cog icon will reveal a details pane that allows us to set pipeline property values.

    • Set the default compute target, this compute resource will apply to all steps in the pipeline.

    • Define pipeline parameters as key value pairs, this can be used to parameterise the pipeline for future runs.

    • Update the pipelines name and provide an optional description.

  4. Pipeline Actions

    These buttons will vary depending on the type and status of the pipeline.

    • Training Pipeline: Submit | Publish | Create Inference Pipeline | Update Inference Pipeline

    • Inference Pipeline: Submit | Deploy

  5. Menu Items

    Autosave | Save | Undo | Redo | Copy | Paste | Delete | Zoom In | Zoom Out | Auto Zoom | Reset Zoom | Multiple Select Mode | Auto Layout | Search in canvas

  6. Last Run Status

    e.g. Not started, Running, Run finished, etc.

  7. Canvas

    Drag and drop modules from the module palette onto the canvas. Create flows of data through your pipeline by connecting the output port of one module to the input port of another.

  8. Properties

    Each module has a set of configurable properties which can be altered from their default values. In order to make these changes, select a module on the canvas to bring it into focus, the properties pane will then appear on the right hand side of the screen. In addition to module specific properties, there are several elements on the details pane that are common throughout all modules, these include:

    • Regenerate Output: True - Force a new run for this module; False - Reuse cached results.

    • Compute Target: Default: Use default compute target; Alternative: Use other compute target.

    • Comment: Optional short description that will display on the graph.

    • Help Documentation: Brief description of the module with a link to further detail on docs.microsoft.com.

3. Azure Machine Learning Concepts

Notebooks
A web-based interface that allows you to create documents that contain live code, visualisations and narrative text.

Experiments
A collection of runs used to validate a user’s hypothesis (i.e. a pipeline execution).

Runs
A pipeline run represents a single trial of an experiment. Run objects are created when you submit a script to train a model.

  • Pipeline Run: Used to manage, check status, and retrieve run details once a pipeline is submitted.

  • Step Run: The execution of a single module within an overall pipeline run.

Models
A model is the result of an Azure Machine Learning training run or some other model training process outside of Azure.

Pipelines
Represents a collection of steps which can be executed as a reusable Azure Machine Learning workflow.

  • Pipeline Draft: Represents a mutable pipeline which can be used to submit runs and create published pipelines.

  • Pipeline Run: Represents a submitted run and can be used to manage, check status, and retrieve run details.

  • Pipeline Endpoint: Represents a Pipeline workflow that can be triggered from a unique endpoint URL.

Compute
A compute target is a designated compute resource where you can run your training script or host your service deployment.

  • Compute Instance: A fully-managed cloud-based workstation for data scientists to get started with Azure Machine Learning Notebooks, CPU or GPU options are available, pre-installed with the most common tools.

  • Compute Cluster: Create a single or multi-node compute cluster for your training or batch workloads.

  • Inference Cluster: Create an AKS cluster or bring your own, to deploy your model on an Azure managed Kubernetes cluster.

  • Attached Compute: Bring your own compute like an HDInsight cluster, a virtual machine, or a Databricks cluster to use as compute with your AzureML workspace.

Datastores
Store connection information to Azure storage services so you can refer to them by name and don’t need to remember the connection information.

Datasets
A reference to data in a Datastore or behind a public web URL. Azure Machine Learning supports two types of datasets: Tabular or File.

Endpoints
The ability to deploy pipeline workflows and machine learning models as web service endpoints so that they can be triggered by external applications with REST calls.

azure_machine_learning_concepts.png

4. Demonstration

 
 

5. Pipeline Workflows

Submit a Pipeline Run
Submit a pipeline run using the compute resources in your Azure Machine Learning workspace.

azure_machine_learning_submit.png
 

Create an Inference Pipeline (Real-time or Batch)
Convert your training pipelines to inference pipelines.

azure_machine_learning_create.png
 

Deploy a Real-time Inference Pipeline to a Real-time Endpoint
Deploy a real-time inference pipeline to a real-time endpoint to make predictions on new data from external applications.

azure_machine_learning_deploy.png
 

Publish a Training or Batch inference pipeline to a Pipeline Endpoint

  • Train or Retrain Models: Publish a training pipeline to reuse a single pipeline to train multiple models while changing parameters and datasets or retrain existing models.

  • Perform Batch Inferencing: Publish a batch inference pipeline to make predictions on new data by using a previously trained model.

azure_machine_learning_publish.png
 

6. Compute Targets

Run Notebook
To execute code from within a Notebook, you must connect to a running compute instance.

aml_compute_notebook.png
 

Submit Pipeline
A pipeline can be submitted against a running compute instance or compute cluster.

aml_compute_run.png
 

Deploy Real-Time Inference Pipeline
A real-time inference pipeline can be deployed to an inference cluster running on the Azure Kubernetes Service (AKS).

aml_compute_inference.png
 

7. Resources

Microsoft Docs: Azure > Machine Learning

Microsoft Docs: Azure > Machine Learning > How-to Guides & Tutorials

Microsoft Updates

Microsoft Learn

YouTube

GitHub Community

/ /
, , , , , , , , ,

Taygan Rifat