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.
-
Enable GPU support when deploying to Azure Container Instances
Azure Container Instances can be configured to provide GPU resources to the container (https://docs.microsoft.com/en-us/azure/container-instances/container-instances-gpu). However, there is no way to ask for GPU resources when deploying ML services via the Azure ML SDK (AciWebservice.deployconfiguration) or Azure CLI (az ml service create aci). Even if the image is built with GPU support enabled (ContainerImage.imageconfiguration(enable_gpu=True,...)), the image does not work properly in ACI because the GPU resources are not present.
Please extend the SDK and CLI to allow specifying the GPU count and SKU when deploying to ACI.
18 votesHi Azure Customer,
Thank you for your feedback. This feature is currently not planned mainly because ACI doesn’t have full GPU support yet. We will follow up with ACI team and expose it when they add full support. At the meantime, we will keep this feedback and vote opening.
Thank you for your understanding.
Regards,
Azure CXP Community -
17 votes
-
Support Network Isolation for Application Insights
Azure ML currently requires you to stand up an Application Insights resource which it uses internally for telemetry information . The Application Insights resource must be on a public endpoint and unlike the rest of the Azure ML components cannot be placed on a Private Link endpoint. Applications Insights supports Azure Monitor Private Link scope, so the resource itself is capable of being placed on a private endpoint.
Please add support to Azure ML to allow for placing the required Application Insights resource on a private endpoint.
16 votes -
Be able to write results to SQL tables through Datasets and Datastores
Currently it doesn't look like it is possible to directly upload the results of a batch Azure ML job to a SQL database that has been registered within the workspace as a datastore. The SQLDataReference class exists, but none of the methods work: https://docs.microsoft.com/en-us/python/api/azureml-core/azureml.data.sql_data_reference.sqldatareference?view=azure-ml-py
It would be great to be able to easily write ML results to SQL databases that the service principal has write access to.
16 votesHi Azure Customer,
Thank you for your feedback here. Your feedback has been forwarded to engineering team and is currently under review.
We will get back to you soon once we have any update.
Thank you for your understanding.
Regards,
Azure CXP Community -
improve new visual interface run execution performance (it's that bad...)
ml studio was already slow, for the new visual is several times worst, even after creating the default 2 node cluster, things that took one minute on ml studio now go up to 8-10 mins on visual preview?understand the tracking happening below, cluster overhead and all that, but the point is, as it is, it's unusable, some screenshots ...
https://twitter.com/rquintino/status/112626349982138777615 votesHi Azure Customer,
Thank you for your feedback, we are aware and investigating it. We will let you know once we have more update.
Regards,
Azure CXP Community -
Add Object Detection Module in Azure ML Designer
Though It is easy to create an image classification model in Azure ML Designer by using DenseNet or ResNet module, it's more convenient if some object detection modules can be provided in designer.
13 votes -
Make http status codes controllable from your scoring file (e.g. move the AMLResponse from Preview)
I am using azureml.core.webservice.Webservice and the python SDK.
Afaik being able to control the HTTP response code of a Webservice is not possible at the moment without AMLResponse which is in Preview and not recommended for production use cases.
This is because the return of the run() function in the scoring file affects only the body of the response of the deployed webservice but cannot change the header.
This means that the scoring file can return an Exception but the response will still have the status code 200. I would like to be able to return 4xx statuses like Bad…13 votesHi Azure Customer,
Thank you for your feedback here. Your feedback has been forwarded to engineering team and is currently under review.
We will get back to you soon once we have any update.
Thank you for your understanding.
Regards,
Azure CXP Community -
Being able to use environments on the compute instance
Currently you can use environments to standardise the compute clusters for distributed training. It would be awesome to be able to use similar logic to configure the compute instances themselves. We usually need to test our script before sending it for distributed training on the compute clusters, but then we need also to have access to the same packages and settings as on the compute cluster!
12 votes -
Allow to modify the bounding boxes during the review of labeled data
We are currently working on an object identification project. We need to label a several number of different objects on one single image.
Currently, we can only choose to accept/reject the entire image when reviewing the labeled data even there is sometimes only one mis-labeled object.
We'd like to be able to adjust the bounding box when it is mis-labeled instead of rejecting everything else.12 votes -
Data Explorer / Kusto as Dataset in AML
We're already using Data Explorer as a way of accessing data from a data lake for a number of other applications. It seems a perfect match for machine learning datasets, with the data already tabularized and often cached. The available alternative is to read the files, but presumably this doubles up on the processing already done by DE.
11 votesHi Azure Customer,
Thank you for your feedback here. Your feedback has been forwarded to engineering team and is currently under review.
We will get back to you soon once we have any update.
Thank you for your understanding.
Regards,
Azure CXP Community -
Need Sample for Azure ml pipeline yaml template
Looking Sample Project for Azure CLI ML Pipeline create from Yaml Template, Please share sample for it.
https://github.com/MicrosoftDocs/pipelines-azureml/tree/master/ml-pipeline-yml
az ml pipeline create -n mypipeline -y mypipeline.yml
https://docs.microsoft.com/en-us/azure/machine-learning/service/reference-pipeline-yaml10 votesHi Azure Customer,
Thank you for your feedback here. Your feedback has been forwarded to engineering team and is currently under review.
We will get back to you soon once we have any update.
Thank you for your understanding.
Regards,
Azure CXP Community -
Please add support for federated learning in AML
More and more users are opting not to send their data back to cloud. In order to respect user privacy and scale the training of our ML models, it would be great if AML would support federated learning.
10 votesHi Azure Customer,
Thank you for your feedback here. Your feedback has been forwarded to engineering team and is currently under review.
We will get back to you soon once we have any update.
Thank you for your understanding.
Regards,
Azure CXP Community -
Ability to run experiments without using SSH port
A lot of enterprise customers have concerns opening SSH port even with VNet enabled and using Azure ML service tag.
We should find a way to run experiments on computes like DSVM without opening up access to SSH port.
10 votesHi Azure Customer,
Thank you for your feedback, we have put it into our roadmap.
Regards,
Azure CXP Community -
AKS AGIC Support for ML Studio
Publishing from ML Studio creates its own Ingress controller, even when you have AGIC Addon/Helm enabled AKS Cluster. It would be nice to have the ML Services deploy to existing AGIC if any before creating a hole to internet.
9 votes -
Add possibility to pass parameters to scoring script when deploying an image
When deploying an image to AML Service, the path of the scoring script is given to the method using the executionscript parameter: ContainerImage.imageconfiguration(execution_script='score.py', ...). However, there is no argument for passing values to script parameters, which are needed in the init() function to do stuff like fetching passwords from Key Vault and getting sample data for the schema.
Please add the possibility to pass parameters to the scoring script when deploying and image to AML Service. For example, when creating a Python script step in an AML Pipeline, we can add these parameters as follows: PythonScriptStep(scriptname='train.py', arguments=['--workspace…
9 votesDear Azure Customer,
Thanks for the feedback. Your feedback is very important to us. We will investigate it and get back to you soon.
Thank you for understanding.
Azure CXP Community -
Enable Forecasting models when lagging features are enabled
Currently, Forecasting models are disabled when user uses lagging features such as target lags and/or rolling windows. They should be added to the roadmap and made available even when lagging features are enabled.
https://docs.microsoft.com/en-us/python/api/azureml-automl-core/azureml.automl.core.shared.constants.supportedmodels.forecasting?view=azure-ml-py
-> 'AutoArima','Prophet','Average','Naive','SeasonalAverage','SeasonalNaive'9 votes -
Add detailed model training progress information in Machine Learning Studio experiment view
At now, the only training progress indicator in experiment view is a time elapsed counter. The idea is to extend number and types of possible counters while the model is being trained. Such information could possibly be: current training iteration, progress percentage, etc.
9 votesHi Azure Customer,
After internal review, this feature request is not currently in the roadmap. We will keep this feedback open and watching for additional upvotes for future priority considerations. Thank you for your feedback again ^^
Regards,
Azure CXP Community -
Increase arbitrarily small metrics document limit (3000b)
Remove the 3000b limit for metrics:
azureml._restclient.exceptions.ServiceException: ServiceException:
Code: 400
Message: (ValidationError) Metric Document is too large, 3195b is larger than the max metric size 3000b.
Details:
Metric Document is too large, 3195b is larger than the max metric size 3000b.This is from running .log_table with only 5 integers (and associated short strings)
9 votes -
Allow ACI and AKS containers to assume a managed identity role
Currently, there's no way for an ACI or AKS-instanced AML program to assume a role. We would like to access other Azure resources with our AML code, so we would like for the containers created to be able to assume an existing role, using the normal procedures.
9 votesHi Azure Customer,
Thank you for your feedback here. Your feedback has been forwarded to engineering team and is currently under review.
We will get back to you soon once we have any update.
Thank you for your understanding.
Regards,
Azure CXP Community -
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
- Don't see your idea?