Machine Learning
Welcome! You can use this site to tell the Azure Machine Learning team what features you would like to see.
Azure Machine Learning enables developers and data scientists to build advanced analytic web services in minutes and hours, eliminating the heavy lifting associated with deploying machine learning in modern data driven applications.
-
Desktop client for AML Studio
The new AML Services offering includes a desktop client, AML Workbench. Please create a desktop client for AML Studio as well. Personally, I find the Web UI for AML Studio a little clumsy sometimes. Ability to use local datasets in AML Studio desktop client would also be nice.
Ideally, AML Studio could be just another feature of the AML Workbench client. No reason to make matters too complicated.
18 votes -
Export Data to storage account that requires secure transfer
I'm trying to use AzureML's "Export Data" to export to a storage account. I got an error (400 Bad Request, similar to the error documented on https://social.msdn.microsoft.com/Forums/en-US/2b7e8220-cd4f-4643-ae1d-1648721b15f4/getting-error-while-trying-to-export-the-data-from-dataset-to-azure-blob?forum=MachineLearning)
After investigation found that the cause was that the storage account had the "Secure transfer require" option enabled. It seems that AzureML's export code needs to be updated to supported encrypted communication with Azure Storage.
17 votes -
Get serious with Datasets! Beat Kaggle
If you are serious about ML Studio, you'd rather include hundreds and hundreds sample datasets. We should see 2-3 samples for each algorithm. Two sets for each dataset needed: 1. real, dirty data 2. the same but clean data. So user could concentrate on cleaning or on the algorithm side of the problem. Datasets should be in categories and with tags... you have fantazy, right?
Samples for different user categories: students, scientists, developers, kids.
Samples for different tasks: study, competitions, science (Yes! Please, include datasets for *real* problems in science)
We should beat Kaggle in this aspect (why not ImageNet?)17 votes -
Extract feature weights of trained neural network model
One advantage I can think of, with this we can observe which attributes are more important. I am sure there are many people out there who don't want the model to be a black box and want to know what exactly goes on behind the curtains.
17 votes -
Multiple output nodes for the regression neural network model
I like to be able to create a regression neural network model that has multiple output nodes. This helps to create an efficient neural net that model multiple related outputs which are dependent on the same group of inputs.
17 votes -
Microsoft Dynamics CRM DataSet Connector
Provide the ability to import data directly from Microsoft Dynamics CRM
17 votesunder review ·
Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
Currently, you can use logic apps CRM connector (https://azure.microsoft.com/en-us/documentation/articles/connectors-create-api-crmonline/) and call Azure ML Batch API from logic apps – https://gallery.cortanaintelligence.com/Tutorial/No-code-Batch-Scoring-and-Retraining-1
-
Add Calibration Module (Reliability Plots, Breier Score, Platt Scaling, Isotonic Regression)
Most learning algos do not produce well calibrated probabilities (typically either overconfident or underconfident) and one cannot directly use the scores in decision making. Two class Boosted Trees are performing well on several other aspects, but the score distribution is poor with bumps close to 0% and 100%. Current scaling transforms do not effectively solve this problem.
17 votes -
Enable grouping of modules on the canvas
Please enable collapsible grouping of modules (perhaps similar to SSIS?)
This would be a functionality that enables the visual grouping of modules that belong together. When chosen to group selected modules (via right click for example), the items are then put into a rectangle that is collapsible to hide all items.
This should have no impact on how the modules work.Reason for this idea is to make the screen easier to navigate when working with large experiments. It could also allow easy copy/pasting of certain modules that belong together.
17 votes -
Number of instances can be displayed as a support feature in decision tree algorithms
In decision tree algorithms, number of instances versus total number of instances forming a particular leaf node can be displayed as a support feature as in R rpart.plot packages so that users get an idea about the important leaf nodes.
15 votes -
More function of OpenCV.
I don't want to re-implement well-known image processing/recognition algorithms. Please implement more features of OpenCV in AzureML, or give us another powerful image tool.
15 votes -
Add anytime package to AML Studio R installed packages
Could you add anytime R package to AML Studio as intalled R package so we do not have to import it as script bundle from a zip file each time we would like to use it?
15 votes -
Webhooks and/or Event Grid Integration
I would like to have the ability for AML to generate an event that external resources can react to when a job completes. It would be nice for AML to call a webhook when a job is complete or publish an event to an Event Grid Topic when a job is complete. Right now, the status of a job has to be polled until the job status is set to complete, and this works, but for long running jobs (30-60 mins) I would rather react to an event instead of polling every second for the duration of the job.
I…
15 votes -
Please document the meaning of some the of Statistical Terms
I have run an N-Gram text processing problem against a Multiclass logistical regression model. The "Evaluate Model" gives me the following results:-
Overall accuracy 0.004621
Average accuracy 0.995271
What does that mean? In short, can you define the following terms: Overall accuracy, Average accuracy, Micro-averaged precision, Macro-averaged precision, Micro-averaged recall, Macro-averaged recall
I can't find a definition of these anywhere on the internet!15 votes -
Add Machine Learning services to use Azure AD Token
When using Machine Learning's API, add a function that can authenticate with Azure AD Token
14 votes -
import from sharepoint
Import from a sharepoint online list would make life about 1000 times easier
14 votes -
Include custom R library in Create R Model task
You currently cannot include your own custom R modules/libraries in the Create R Model task. The functionality is already available for the Execute R Script (see https://msdn.microsoft.com/en-us/library/azure/dn905952.aspx). You should be able to include it the same way:
install.packages("src/libraries/customLibrary.zip", lib='.', repos=NULL, verbose=T)
Or
source("src/customLibrary.R")14 votes -
Add Bayesian Network Algorithm and Adaboost
Having these two algorithm in Azure will make life easier.
13 votes -
Need Advanced Time Series Models in AZURE ML like ARIMA, ARIMAZ, SARIMAX etc. for various industrial applications
Time series models have numerous industrial applications and most widely used models in R & Python are ARIMA, ARIMAX, SARIMAX etc.
Request to add all advanced time series models in Azure ML.13 votes -
Data import from onedrive
Do we have the feature of importing data to Azure ML from one drive? is this feature being considered
13 votesunder review ·Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
It is certainly something in consideration.
-
Save Python model objects as trained models
It would be nice to be able to save fitted objects from sklearn as trained models (similar to what is now doable with R scripts) without having to go through the process of pickling, reading a dataset, and unpickling.
13 votesplanned ·Adminmluservoice
(Program Manager, Microsoft Azure Machine Learning, Microsoft Azure)
responded
This feature is on our roadmap.
- Don't see your idea?