Big Data and Machine Learning on AWS
- 1. Big Data and Machine Learning on AWS AWS User Groups of Florida April 2018 Patrick Hannah, VP of Engineering, CloudHesive
- 2. About Me • Who am I? • What’s my background? • What do I hope to get out of the presentation? • How am I using AWS?
- 3. About CloudHesive • Professional Services – Assessment (Current environment, datacenter or cloud footprint) – Strategy (Getting to the future state) – Migration (Environment-to-cloud, Datacenter-to-cloud) – Implementation (Point solutions) – Support (Break/fix and ongoing enhancement) • DevOps Services – Assessment – Strategy – Implementation (Point solutions) – Management (Supporting infrastructure, solutions or ongoing enhancement) – Support (Break/fix and ongoing enhancement) • Managed Security Services (SecOps) – Encryption as a Service (EaaS) – encryption at rest and in flight – End Point Security as a Service – Threat Management – SOC II Type 2 Validated • Next Generation Managed Services – Leveraging our Professional, DevOps and Managed Security Services – Single payer billing – Intelligent operations and automation – AWS Audited
- 4. What are we going to talk about? • Big Data and Machine Learning • Common Use Cases • AWS Services in support of Big Data and Machine Learning • Demos • Conclusion
- 5. Let’s define Big Data and Machine Learning • From Wikipedia: – Big data is data sets that are so voluminous and complex that traditional data processing application software are inadequate to deal with them
- 6. Let’s define the driver
- 7. Let’s talk about some of it’s applications • Research – Grid/HPC Computing (the original cloud) – Initiator of open source projects – Enabler and enabled by Public Cloud – AWS Just Announced OpenData Registry: https://registry.opendata.aws/ • Business Operations – ERP – Data Warehouses – Business Intelligence – Business Systems • Applied – Every Major Industry – {Dev|Sec|Ops} – Products (b2b, b2c)
- 8. Let’s talk about its characteristics • Lifecycle driven – Collect – Store – Process/Analyze – Consume • Generation – Batch – Streaming • Format – Text – Images – Audio – Video
- 14. Machine Learning - Layers of Abstraction • Application Specific (All Levels) – Amazon Comprehend – Amazon Lex – Amazon Polly – Rekognition – Amazon Transcribe – Amazon Translate • AWS Machine Learning (Beginner) • AWS SageMaker (Intermediate) – AWS DeepLens • Deep Learning Learning AMIs (Advanced)
- 15. Overview of Machine Learning • What is Machine Learning? – A subfield of computer science that evolved from the study of pattern recognition and computational learning theory in artificial intelligence. • What is AWS Machine Learning? – A platform that allows software developers to build and train predictive applications and host those applications in a scalable AWS cloud solution.
- 16. Key Terms for AWS Machine Learning • Datasources – Contain metadata associated with data inputs to Amazon ML (your sample data) • ML models – Generate predictions using the patterns extracted from the input data • Evaluations – Measure the quality of ML models • Batch predictions – Asynchronously generate predictions for multiple input data observations • Real-time predictions – Synchronously generate predictions for individual data observations
- 17. What problem are we trying to solve? • Alerting on event data (which we will describe on the next slide) is based on traditional mechanisms: • Threshold crossed > alert • Pattern matched > alert • These mechanisms are consistent, until an outlier comes along. • When an outlier comes along, we need to manually evaluate it • When it comes along again, we add an exception for it • Why not leverage Machine Learning to do this for us?
- 18. Get our event data in one place • Collect from Disparate Systems – Structured Data (Key/Value, Time/Series) • CPU, Memory, Storage, IO, Bandwidth – Unstructured Data (Logs) • Windows Event Logs • Linux /var/log • E-Mail • Third party systems • Normalize it (into a common format) • Push it (to a stream)
- 19. Evaluate and Action on it with Machine Learning • Once a threshold has been crossed, but before we take action on it pass it to AWS Machine Learning (via Kinesis) • AWS Machine Learning uses the previously designated Model to determine the likelihood of the event being a false positive • If Machine Learning determines it’s a false positive, it gets logged in the event stream • If Machine Learning determines it’s an actionable event, it is forwarded on to our alert system (via SNS)
- 20. Why use Machine Learning to Solve This Problem? • Consistency – No longer is a human making judgement call (which will vary from person to person) – No longer is a human taking manual action to whitelist/blacklist/filter the event (which may be done inconsistently) • Cost Savings – The cost of this making the judgement call (in distraction, time and errors) outweighs the cost of the service – At $0.0001 per prediction, assuming 1% of events are false positives your cost for automatically detecting a false positive is $0.01 (1 Cent) versus the cost of paying a human to manually detect a false positive
- 21. Conclusion • AWS provides a number of services to support your Big Data and Machine Learning needs • Getting started on AWS is easy; with the free tier, you can experiment with a number of services without incurring significant cost. • Adoption of AWS in your organization can be as easy or as hard as you want to make it; start simple and iterate.
- 22. Demos • Demos
- 23. Further Learning • Getting Started: https://aws.amazon.com/getting-started • General Reference: http://docs.aws.amazon.com/general/latest/gr • Global Infrastructure: https://aws.amazon.com/about-aws/global-infrastructure/ • FAQs: https://aws.amazon.com/faqs • Documentation: https://aws.amazon.com/documentation/ • Architecture: https://aws.amazon.com/architecture • Whitepapers: https://aws.amazon.com/whitepapers • Security: https://aws.amazon.com/security • Blog: https://aws.amazon.com/blogs • Service Specific Pages: https://aws.amazon.com/service • AWS Answers: https://aws.amazon.com/answers/ • AWS Knowledge Center: https://aws.amazon.com/premiumsupport/knowledge-center/ • SlideShare: http://www.slideshare.net/AmazonWebServices • Github: https://github.com/aws and https://github.com/awslabs
Editor's Notes
- #8: AMAZON DOT COM!!! Agriculture, Forestry, Fishing and Hunting Mining, Quarrying, and Oil and Gas Extraction Utilities Construction Manufacturing Wholesale Trade (41 in Canada,[3] 42 in the United States[2]) Retail Trade Transportation and Warehousing Information Finance and Insurance Real Estate and Rental and Leasing Professional, Scientific, and Technical Services Management of Companies and Enterprises Administrative and Support and Waste Management and Remediation Services Educational Services Health Care and Social Assistance Arts, Entertainment, and Recreation Accommodation and Food Services Other Services (except Public Administration) Public Administration
- #10: From re:Invent 2017
- #11: From re:Invent 2017 +MQ +DMS +Kinesis Video
- #12: From re:Invent 2017 +CloudSearch
- #13: From re:Invent 2017
- #14: +Quicksight +Glue +Data Pipeline
- #16: Machine Learning: Subset of Predictive Analytics Various techniques/approaches that I won’t get into Numerous software products available Examples: Good Example: Marketing, Fraud Detection, Risk How Target Knew a High School Girl Was Pregnant Before Her Parents Did (http://techland.time.com/2012/02/17/how-target-knew-a-high-school-girl-was-pregnant-before-her-parents/) Machine Learning is consistent and not subject to human error but garbage in = garbage out. Like any piece of technology you are giving up control for perceived benefits (I want to see every event and assess it’s validity versus letting Machine Learning do it for me (cite example of me using Erlang for capacity planning) AWS’ Machine Learning: Similar characteristics to other AWS Services (Cloud: Managed, Abstracted) Once key terms are understood, it’s easy to get started (I’m a great example of this) Don’t need to pick software, stand up EC2 instances, install it, configure it, learn it
- #17: Our interest is in real-time predictions (100 ms, being real time) Output (target) is binary (1/0), multiclass (a,b,c) or prediction (3.141) Well suited for situations where manual effort or logic is too complex
- #20: Based on https://github.com/awslabs/machine-learning-samples/tree/master/social-media