Update: Microsoft will be moving away from UserVoice sites on a product-by-product basis throughout the 2021 calendar year. We will leverage 1st party solutions for customer feedback. Learn more here.

Machine Learning

Welcome! The Azure Machine Learning team invites you to share and vote for features to help you build, manage, and deploy custom machine learning models.

Have a technical question, or want to learn more? Please visit our documentation, MSDN forum or StackOverflow.


How can we improve Microsoft Azure Machine Learning?

You've used all your votes and won't be able to post a new idea, but you can still search and comment on existing ideas.

There are two ways to get more votes:

  • When an admin closes an idea you've voted on, you'll get your votes back from that idea.
  • You can remove your votes from an open idea you support.
  • To see ideas you have already voted on, select the "My feedback" filter and select "My open ideas".
(thinking…)

Enter your idea and we'll search to see if someone has already suggested it.

If a similar idea already exists, you can support and comment on it.

If it doesn't exist, you can post your idea so others can support it.

Enter your idea and we'll search to see if someone has already suggested it.

  1. Change resource group for LB, Public IP, and NSG of Compute Instance

    Lb, public IP, and NSG, which are created when a compute instance is created behind a virtual network, are created in the same resource group as the virtual network.

    If the virtual network and AML workspace are in different resource groups, it can be difficult to understand the cost of AML.

    For the convenience of cost control, want to make improvements that LB, Public IP, and NSG are created in the same resource groups as the AML workspace or the use of a tag common to the AML workspace.

    8 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →

  2. Don't have invisible meta data from scoring, that cannot be replaced

    It turns out that the Score activity in the Designer, adds some invisible meta data, called Score, that if you do any Arithmetic work on, will remove that metadata on the column, and make the Evaluator stop working. Just spent about 14 tech support calls tracking this down.

    Here is what he said:
    This is because the Evaluate Model module is expecting a ‘Scored Dataset’, which means the Score columns (predicted column, Scored Labels, Scored Probabilities) all have metadata assigned to them so that their ‘Feature type’ is Numeric/String Score instead of Numeric/String Feature. Passing the scored dataset into ‘Apply…

    8 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    1 comment  ·  Flag idea as inappropriate…  ·  Admin →

  3. Add PowerShell to Azure Notebooks

    Please add and improve the kernel for Windows PowerShell/Powershell Core/6/7 to Azure Notebooks.
    PowerShell should be treated as first citizen language and there are a lot of usecases for teaching with Azure Notebooks.
    Alternatively provide an option to load custom kernels.

    7 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →

  4. Enable Datastores Service in AML to access the default Storage Account using "Resource Instances" in the integrated Firewall of the SA

    Resource instances by adding Microsoft.MachineLearningServices in the integrated Firewall of the Storage Account when it's behind a Vnet enable us to authorize the workspace to write experiments output, models and logs.

    Activating the integrated firewall however blocking the Datastores of the workspace to access the storage account (such as Designer or Explore DS) in scenarios where the private endpoint is not applicable to be implemented.

    It will be a huge advantage if the Datastore uses the same Resource Instance as (or some other feature) in order to access the default storage account of the workspace privatly without the need of…

    7 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →

  5. Support Python 2.x

    Please support Python 2.x as many frameworks still use it.

    7 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    1 comment  ·  Flag idea as inappropriate…  ·  Admin →

  6. Allow attachment of a VM to AML to private IP address

    Please allow attachment of a virtual machine to machine learning studio by using private ip address of a VM. Due to company security restrictions it is not possible to use public ip address. Also from security point of view it doesn't make sense to use public ip only.

    7 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →

  7. Add segmentation to data labeling tool

    You have bounding box labeling but not anything to help with segmentation.

    7 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →

  8. Migration from Azure Machine Learning Studio to Visual Interface

    Ideally this should be done automatically for us, but I'd be happy with an export/import flow that lets us migrate manually.

    7 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    1 comment  ·  Flag idea as inappropriate…  ·  Admin →
    planned  ·  Adminazurecxpuservoice (Product Manager, Microsoft Azure) responded

    Hi Azure Customer,

    Our plan is to support convert the Studio experiment to visual interface experiment. Model and web service are currently out of scope. Thank you for your feedback and please keep eyes on it.

    Regards,
    Azure CXP Community

  9. Enable managed identity for compute instance

    For compute clusters we are able to use managed identities but not for compute instances. And there are functions for the Azure ML SDK that need authentication, especially the Workspace method, which by default assumes an interactive authentication method: https://docs.microsoft.com/en-us/python/api/azureml-core/azureml.core.workspace.workspace?view=azure-ml-py#from-config-path-none--auth-none---logger-none---file-name-none-

    6 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →

  10. Default dependency improvement request for ReinforcementLearningEstimator instances

    The following settings of script parameters are not available in the default environment, and you must install tensorflow==2.1.0 and tensorflow_probability==0.9.0 using the pip_packages parameter.

    training_algorithm = 'SAC'
    rl_environment = 'Pendulum-v0'

    Since SAC is a new and very powerful algorithm, it is in high customer demand, so please make it available in the default dependency.

    6 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →

  11. Request to make the Ray Library 1.0.1 or later available in the ReinforcementLearningEstimator class

    Let me ask you about the parameter settings for the ReinforceLearningEstimator class at the link below.

    https://docs.microsoft.com/en-us/azure/machine-learning/how-to-use-reinforcement-learning#define-a-worker-configuration

    Pip packages we will use for both head and worker

    pip_packages=["ray[rllib]==0.8.3"] # Latest version of Ray has fixes for isses related to object transfers

    Setting the pip_packages parameter of the ReinforcementLearningEstimator class to ['ray[rllib]==1.0.1'] or later will result in a "no such option: --redis-port" error. I suppose this is because the Ray 1.0.1 and later doesn't support "--redis-port" parameter and the implement of ReinforcementLearningEstimator class is fixed to use "ray start --head --redis-port_6379 ~".

    I know it's a preview feature now, but please…

    6 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →

  12. Monitor AML Cluster performace

    Add to AML Python SDK functions to monitor AML compute cluster's load/performance (CPU, Mem, Disk I/O, Network, GPU).

    6 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →

  13. Link Dataset, Experiment, Image, Model, Deployement

    As part of managing the whole Data Science Process within Azure ML, is there a way to link a Dataset, an Experiment and related produced Image, Model and Deployment, apart from description.

    6 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    2 comments  ·  Flag idea as inappropriate…  ·  Admin →

  14. Visualize support for Designer's Train Model module

    Ml Studio (classic) was able to visualize the result, but Designer doesn't seem to be able to do it. To view the model parameters and feature weights, it's better to support Visualize.

    6 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →

  15. Support python 3.7

    Support running Python 3.7 scripts

    5 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    1 comment  ·  Flag idea as inappropriate…  ·  Admin →

  16. We have to be able to export experiment in AML Studio so that it can be imported later with another account

    I developed an experiment in AML Studio and deployed it as a web service.

    I would really appreciate it if I can I export it so that it can be imported later with another Azure Account and it can be redeployed again (in case that i don't have my account anymore or i delete my current experiment)

    5 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    1 comment  ·  Flag idea as inappropriate…  ·  Admin →

  17. Document azure devops machine learning tasks

    To create azure devops pipelines implementing MLOps(e.g. [1], you have created pipeline tasks which can be executed in the devops pipeline, e.g ms-air-aiagility.vss-services-azureml.azureml-restApi-task.MLPublishedPipelineRestAPITask@0 and others. There clearly mirror the CLI and API, but they provide no documentation. They should.

    1: https://github.com/microsoft/MLOpsPython/blob/master/.pipelines/diabetes_regression-ci.yml

    5 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    1 comment  ·  Flag idea as inappropriate…  ·  Admin →

  18. Remove AKS minimum core restriction ("cluster_purpose")

    If I want to attach an AKS cluster in development environments that don't meet the 4 agents / 12 core requirements, I need to set the "clusterpurpose = AksCompute.ClusterPurpose.DEVTEST" parameter via the Python API. However, this parameter is not supported in ARM templates and also not in the Azure CLI extension, so this complicates our automation strategy.

    The minimum requirements for running an AKS cluster in production are well documented, so it shouldn't be up to the "Machine Learning service" to enforce any restrictions.

    Having this "cluster_purpose" parameter is just an unnecessary complication, and should therefore be removed…

    5 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →

  19. Deploy same name service in Azure machine learning workspace service.

    In Aml workspace, we are allowed to add multiple Compute assets for a type. Ideally ml services should able be deploy on both AKS cluster but due to some reason, same named service can't be maintain in Deployment asserts. For example, I can have two AKS Compute resource in a workspace named as AKS1 and AKS2 and service named as 'intelligentAgent'. This service can be deploy in AKS1 but I have trouble deploying same service in AKS2.

    I need to do this to maintain services for BCP and disaster recovery.

    Could you please allow same named services on different AKS…

    5 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    1 comment  ·  Flag idea as inappropriate…  ·  Admin →

  20. Allow for filtering to experiment / error type granularity in Azure Monitoring

    Allow for filtering to experiment granularity in Azure Monitoring.

    Currently metrics only send value 1 if send to Azure Monitoring e.g. if an experiment run succeeds. During experimentation the larger share of experiments will fail, so we need to understand both the error type and the specific experiment at least.

    5 votes
    Vote

    We're glad you're here

    Please sign in to leave feedback
    Signed in as (Sign out)
    Close
    Close
    You have left! (?) (thinking…)
    0 comments  ·  Flag idea as inappropriate…  ·  Admin →
1 3 5 … 12 13
  • Don't see your idea?

Machine Learning

Categories

Feedback and Knowledge Base

(thinking…)
Hosted by UserVoice