Azure Machine Learning
Download as PPTX, PDF2 likes923 views
This document discusses the evolution of machine learning tools and services in the cloud, specifically on Microsoft Azure. It provides examples of machine learning frameworks, runtimes, and packages available over time on Azure including Azure ML (2015) and the Microsoft Cognitive Toolkit (CNTK) (2015). It also mentions the availability of GPU resources on Azure starting in 2016 and limitations to consider for the Azure ML service including restrictions on programming languages and a lack of debugging capabilities.
Azure Machine Learning
- 1. Machine Learning with Microsoft Azure Dmitry Petukhov, ML/DS Preacher, Coffee Addicted && Machine Intelligence Researcher @ OpenWay
- 2. Storage Resource Management ML Framework Execution Engine Local OS Local Disc RRuntime YetAnotherRuntime ML packages HDFS YARN MapReduce Mahout HDFS / S3 YARN / Apache Mesos Apache Spark MLlib HDFS / S3 YARN / Apache Mesos PySpark / SparkR Apache Spark Local PC Hybrid Model Cluster (on-premises/on-demand) Low HighCost of deployment/ownership Distributed FS Dark Magic… ML as a Service DS Tools Infrastructure for Data Scientists ML Packages
- 3. ML runtime as a Service • Feb 2015: Azure ML • Apr 2015: Amazon ML • Oct 2015: Google Cloud ML Engine Deep Learning algorithms in Open Source • Dec 2015: The Microsoft Cognitive Toolkit (CNTK) • Nov 2015: Microsoft Distributed Machine Learning Toolkit • Nov 2015: TensorFlow (Google) • May 2016: Amazon DSSTNE Deep Learning models as a Services • Microsoft Cognitive Services • Amazon Rekognition • Google APIs: Natural Language, Speech API, Translation API, etc. GPU on demand • Sep 2016: private preview in Azure • Oct 2017: Amazon update its GPU-instances • March 2017: Google GPU features 2015 2016 2017 Machine Learning Tools Evolution in Cloud
- 5. Data Azure Machine Learning Consumers Cloud storage Business problem Modeling Business valueDeployment Model REST API Manage Local storage REST API Reference: Microsoft Data Amp 2017 ML Studio Web IDE + ML runtime ML Web Services ML-model publication Azure Marketplace Cortana Gallery Data Model
- 6. Bitcoin Secret Demo library(Quandl) library(dplyr) library(forecast) dt <- Quandl("BCHARTS/BITBAYUSD", api_key = ****** ) %>% select(Date, Close) dt.train <- dt %>% filter(Date < max(Date) - months(2)) dt.test <- dt %>% filter(Date > max(dt.train$Date)) timeseries <- ts(dt.train$Close, frequency = 1) model <- ets(timeseries)
- 7. Restrictions Legislative Restrictions International & local Azure Platform Limits Max storage volume per account, etc. Azure ML Service Limits Black box No debug No Scala, or C++, or C# No your own “right” algorithms No Deep Learning See also Service Tiers Limitations Azure Machine Learning: Limits
- 8. © 2017, Dmitry Petukhov. CC BY-SA 4.0 license. Microsoft and other product names are or may be registered trademarks and/or trademarks in the U.S. and/or other countries. Thank you!
- 9. Q&A Now or later (use contacts below) Ping me Habr: @codezombie All contacts: http://0xCode.in/author
Editor's Notes
- #4: Cutting Edge 2015: ML Runtime as a Service Feb 2015: Azure ML Apr 2016: Amazon ML Oct 2015: Google Cloud ML Engine 2015-2016: Deep Learning algorithms in Open Source Dec 2015: The Microsoft Cognitive Toolkit (CNTK) Nov 2015: Microsoft Distributed Machine Learning Toolkit http://www.dmtk.io Nov 2015: TensorFlow https://github.com/tensorflow/tensorflow/ May 2016: Amazon DSSTNE https://github.com/amznlabs/amazon-dsstne 2016: Cognitive Services Microsoft Cognitive Services Amazon Rekognition Google APIs: Natural Language, Speech API, Translation API, etc. 2016-2017: GPU on demand Sep 2016: private preview in Azure Oct 2017: Amazon update its GPU-instances March 2017: Google
- #5: Source: https://blogs.technet.microsoft.com/dataplatforminsider/2017/04/19/delivering-ai-with-data-the-next-generation-of-microsofts-data-platform/
- #9: (c) 2017, Dmitry Petukhov. CC BY-SA 4.0 license.