Important update for Azure Machine Learning
úterý 9. prosince 2014
The December 8, 2014 release of Azure Machine Learning includes a major version upgrade of the modules and machine learning framework. For more information, please see Major Version Upgrade on 12/8/14 May Require User Action. New features and improvements
- New modules that are available:
- Two-Class Locally-Deep Support Vector Machine module
- Pre-trained Cascade Image Classifier module for face detection, using OpenCV library.
- Image Reader module that uses OpenCV library to read images from Azure Blob storage
- Sweep Parameters module supports cross-validation mode, in addition to the train-test model. The test input port becomes optional. If it is left unconnected, cross-validation mode is used.
- Sweep Parameters module orders the output table by the selected evaluation metric.
- Decision Forest Regression returns standard deviation for scored labels.
- Timeouts for Reader and Writer modules is increased. The issue with 15-minute default timeout for Hive ingress has been addressed by increasing the timeout to 24 hours.
- In the Filter-Based Feature Selection module, Kendall, Pearson, and Spearman correlations use conditional mean expectation value in cases when one column is numeric and other one is categorical.
- Evaluate Recommender module can automatically infer the type of recommendation based on column names.
- Removed the Ready to Deploy to Production button from Studio. This enables publishing the Web Service from an experiment as a "default end point" without the need to take any additional action to deploy it to production.
- The published "default end point" is shown in the Azure portal under Web Services. There, a new end point can be created from the default end point, if needed (for example, to share with a customer). In the Studio, the additional end point count will show as "0" until a new end point is created in the Azure portal. (Currently, this feature is limited to adding one end point in addition to the default.)
- When you create a new experiment, the experiment tile names in the gallery of sample experiments wrap so that you see the entire experiment name.
- When you create or edit an experiment, you'll notice that the resize area between the experiment canvas and the module palette is increased.