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 re-training of the model via WebService / API call
Many live-data scenarios such as algorithmic trading require frequent re-training of the model when new live data comes in. In this case the model needs to be re-trained based on the new data window. For these kinds of applications to be feasible with Azure ML, it's required that the re-training / publishing of the model can be automated. It's not necessary that this is solved using a WebService call, it could be done using some exposed API call from .NET, etc.
174 votescompleted ·
Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
This is now available. Please refer to the documentation here: http://azure.microsoft.com/en-us/documentation/articles/machine-learning-retrain-models-programmatically/
-
Expose API for model building / publishing automation
When offering ML Prediction as Service to our customers it can be crucial to automate model creation, as different models need to be generated for different customers. It'd probably be enough to enable importing R scripts via an API or WebService call to solve this.
164 votescompleted ·Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
Please see http://aka.ms/amlps
-
Save R-Script as trained model
Currently only built-in models in Azure ML can be saved and published. We've built a model in R and want to publish it as well but Azure ML does not support it right now. I wonder if Azure ML can somehow put some rules on the input/output of the R-Script, so when a script follows these rules, it can be saved and published like any other built-in models. This could increase the extensibility and flexibility of Azure ML pretty much.
120 votescompleted ·Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
You can use create R model described here – https://msdn.microsoft.com/en-us/library/azure/dn955435.aspx
-
Enable and Disable Steps in an Experiment
When experimenting, the ability to disable steps makes it easier to try different approaches. For example, I wanted to have different sets of columns as inputs to a regression. I have 4-5 different Project Column steps with different fields and swap the output to the inputs to select. However, I'd love to just disable them on a right-click menu item.
111 votesHi,
The Run Selected feature satisfy your requirement. Please select the modules you want to run and just click the “Run Selected” button.
Thank you,
Azure CXP Community -
Deployment of ML experiments programmatically
Presently, the ML workspace and experiments needs to be manually created. There is no capability to create new workspace and experiments using some scripting like, C# or PowerShell. It will be good, if we can take backup or generate deployable files from existing ML experiments which in turn can be used for deploying it to another/same subscription with same/different name.
94 votes -
Add Package ‘DMwR’
I need to use lofactor function. This function obtain local outlier factors using the LOF algorithm.
90 votescompleted ·Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
We have released support for Microsoft R Open 3.2.2 as part of execute-R module. That version has this package
-
Run only part of a model
When building a model, running it takes quite some time and while working on it, I may not have fully specified all elements of my model as I create the outline of the proces.
Therefore I would like to be able to run the model up to a certain step, ignoring all elements after the selected block. This would save compute and wait-time and would fit my way of working nicely.
61 votescompleted ·Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
Please see the “Run Selected” feature in the Azure ML Studio which allows you to run only a group of selected modules within a graph.
-
Show variable importance after experiment runs.
One of the more interesting things to observe when building a model is seeing what variables were determined to be most important.
Allowing us to see various variable weights (p values for linear regression, importance in tree based models, etc.) would make experiments more insightful.
59 votescompleted ·Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
Please see the Permutation Feature Importance module.
-
Provide support to upload compressed datasets
It would be great if we are allowed to upload a ZIP file containing a TXT or CSV dataset. This will save on bandwidth and time for end users. Currently I can upload a ZIP file but I am unable to visualize that as a normal dataset.
58 votescompleted ·Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
You can upload a ZIP file and then use an Unpack ZIP module to extract the dataset.
-
Support online learning for recommender
It looks like Matchbox recommender doesn't support online learning at the moment. This makes impossible to provide recommendation for the new users based on their profile/preferences.
44 votescompleted ·Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
We have support for Vowpal Wabbit which provides online learning. We look for more opportunities based on customer scenarios and ask. We also support retraining capability which can be programatically triggered for ongoing model updates
-
Deploy a web service to AKS without a public endpoint
It seems like the Machine Learning service currently always uses a public load balancer with a public IP when deploying a web service to AKS.
We do not want to expose our web services to the public as that is a security risk - even with the Authorization keys. Also, even if we had to expose it, we'd prefer to use our own centralized ingress system to control the traffic coming into the cluster (e.g. via Azure Application Gateway v2).
We therefore would like to be able to choose one of the following additional methods when connecting to AKS/when deploying…
41 votesThis scenario is supported now:
https://docs.microsoft.com/en-us/azure/machine-learning/how-to-enable-virtual-network#use-private-ips-with-azure-kubernetes-service.Thanks.
-
It would be great to have access to all the Python as well as R
I'm not familiar with R and don't wish to invest in learning it, when Python packages such as Scipy, Numpy, Pandas and Scikit-Learn meet most of my needs. I wish I could combine the power of these with Azure ML's slick interface, and easy-to-deploy APIs
41 votescompleted ·Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
This is now supported – here is the sample experiment – http://gallery.azureml.net/Details/978fa971bca845f39ee07fbbec23143d
-
Recommendation API tutorial
Please provide a tutorial on how to recommend movies etc to someone using Azure ML based off of previous customer history (cold start) and your own history.
Similar to http://blogs.technet.com/b/machinelearning/archive/2014/07/09/recommendations-everywhere.aspx
40 votescompleted ·Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
here is the steps for recommendation system – http://gallery.azureml.net/Details/3a02931f94114f47b4512dd9179b515e and here is the video – http://gallery.azureml.net/Details/3a02931f94114f47b4512dd9179b515e
-
Can you please add xgboost library to the existing list.
Can you please add xgboost library to the existing list.
40 votescompleted ·Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
We have added support for MRO 3.2.2 for execute R module which has this package
-
Add R plugin in Visual Studio for syntax highlight and turns VS in R IDE
The plugin exist for R Revolution, just add it like Python plugin for V IDE.
39 votes -
Add a folder structure to the my experiments and webservices tabs
Please add the ability to create sub-folders and move around items in the 'My experiments' and 'webservices' tabs, to help us organise our workspaces better.
30 votescompleted ·Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
Please see the “Project” feature.
-
Machine Learning Workspaces
Looking though the new portal, there are many cool things. However, I haven't been able to find a way to access my AzureML workspaces from the new portal
29 votescompleted ·Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
Azure ML workspace can now be created in Ibiza
-
28 votescompleted ·Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
Available in west europe
-
Please upgrade R version in Azure ML Studio.
Please upgrade R version in Azure ML Studio. It is very inconvenience use the latest R packages.
23 votescompleted ·Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
We have added MRO 3.2.2 support.
-
Get machine learning experiments under source control
It should be awesome if we could export (test) experiments and put them under source control in for example TFS, while also being able to import these experiments in a new (production) workspace.
15 votescompleted ·Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
We have added support for import and export web service definition and experiment JSON through PowerShell and store in version control repositary
Web Service – msdn.microsoft.com/en-us/library/azure/mt76..
Experiment – github.com/hning86/azuremlps
- Don't see your idea?