חידושים מעניינים בנושא General Availability release of Azure Machine Learning RRS feed

  • דיון כללי

  • Larry Koch ממיקרוסופט הוציא את ההודעה הבא:

    On February 18th, 2015, Microsoft announced the General Availability release of Azure Machine Learning--now the most comprehensive service in the cloud for advanced analytics.  
     There are many exciting new features and improvements that have become available with this release:
    Azure Machine Learning Gallery (Preview)
    We are introducing a new experience to Discover, Learn and Share Azure Machine Learning experiments.
    Browse, search and share experiments authored by others as well as sample experiments created by Microsoft data scientists.
    Copy experiments from the Gallery into your ML Studio workspace using “Open in Studio” link from the experiment Details Page.
    For additional details and FAQ, check out: Learn how to contribute
    Machine Learning Studio
    Go through Guided Tour and Tutorial Experiment to quickly find your way around the ML Studio environment.
    Contribute to the Gallery from within the Studio using “Publish from Gallery” command button.
    Ability to delete uploaded Datasets and saved Trained Models.
    Explore numerous new Samples experiments to cover newly released modules and previously not covered functionalities
    Added data science templates (published as multi-step sample experiments) for domains including Text classification, Fraud Detection, and Retail Forecasting.
    Python Support
    Execute custom Python script in ML Studio using Execute Python Script module.
    Download/upload dataset using the Azure ML Python Client Libraries found on Github.
    Generate Python code for accessing datasets from ML Studio.
    New Modules
    Create R Learner for creating R based machine learning models.
    Built Count Table for building count tables for frequencies of feature variables that appear for positive and negative samples. Computation can be done on an HDInsight cluster to featurize large datasets.
    Count Featurizer for using count tables to construct count and frequency based feature vectors.
    Fast Forest Regression for returning predictions with quantile bounds.
    Execute Python Script module for executing Python scripts and visualizing data.
    Principal Component Analysis for dimensionality reduction.
    Web Services (APIs and Operationalization):
    ML Studio
    Web Service Input and Output modules replace the previous input and output ports.
    Auto-generate Scoring experiment from Training experiment.
    Auto-prepare a Training experiment to be published as a web service.
    Toggle between Web service flow view and experiment flow view.
    Add multiple input and output modules for publishing web services.
    Consume Request/Response Web API from downloadable Excel workbook.
    Azure Portal: Multiple endpoint support
    Add/remove Web service endpoints from the Azure Portal Web service page and set the throttling level.
    Web Service APIs
    Programmatic retrain models published to web services.
    REST API for Web service management operations
    Web services can be published with multiple outputs to allow returning of multiple results from the same call for scoring, model evaluation, etc.
    Allow graphic device port to be output from a web service.
    Support for mini-batch (multi row) input requests to the web service
    Automatic web service graph optimization - avoiding execution of modules that do not contribute to results calculation in web services
    Batch Execution Service (BES) improvements:
    To enable fully idempotent job execution and reliable service clients, the BES job submission will now be a two-step process:
    Returning of job status and errors as descriptive text
    Web service failures are now reported to the user in a descriptive way
    Web Service diagnostic and profiling can be configured on user’s own blob storage.
    Ability to scale web service up to 200 concurrent calls.
    Deprecated Modules
    Apply Quantization Function. Replaced by Apply Transformation.
    Linear Discriminant Analysis. Replaced by Fisher Linear Discriminant analysis.
    Missing Value Scrubber. Replaced by Clean Missing Data.
    Quantize. Replaced by Quantize Data.
    Cloud Data Science Process
    Created an end to end cloud data science process map that walks a customer from getting data from various sources into Azure, getting it ready for Azure Machine Learning and using Azure Machine Learning to create and operationalize models for consumption by end user applications.
    Created guides on moving data into SQL Server in Azure and Azure Blob Storage using various tools.
    Created guides on processing and sampling data to get it ready for Azure Machine Learning using SQL and Python.
    Created end to end walkthroughs that use the cloud data science process with the NYC Taxi Trip dataset ( as an example.

    ההודעה המקורית כאן:

    signature   Ronen Ariely
     [Personal Site]    [Blog]    [Facebook]

    שבת 28 פברואר 2015 07:29
    מנחה דיון

כל התגובות

  • תודה על השיתוף

    מיקרוסופט מציעה שירות זה ללא תשלום, למטרת סיוע למשתמשים והעשרת הידע הקשור בטכנולוגיות ובמוצרים של מיקרוסופט. תוכן זה מתפרסם כפי שהוא והוא אינו מעיד על כל אחריות מצד מיקרוסופט.

    יום רביעי 18 מרץ 2015 10:09