What you missed in analytics this week

Just in case you couldn't keep up with everything published on the internet last week, here are the four most interesting analytics reads I found last week - and one that just pushed my buttons...

• The 5-Minute Guide to Understanding the Significance of Apache Spark.  Love this.  Finally someone starts to write explanatory stuff about the Hadoop ecosystem in plan English.  Nitin Bandugula, I salute you sir.


• Keen IO targets enterprises with embedded analytics engine.  This company is new to me, but I'll be learning more later this week.  Although it doesn't actually promote itself as such, I'm thinking this might be an Industrial IoT analytics platform in disguise.  I notice that (a) John Deere is listed as a customer, and (b) the solution is designed to be embedded, so an interesting option for anyone who might want to roll their own IIoT solution.


• Oh, I love this too.  Machine learning is particularly good when applied to problems where the logic is nigh on impossible to distill and code.  But, that very lack of transparency makes it tricky to wrap your head around it and understand how it actually works.  I hope I did a decent job of uncovering some of the mystery in my last blog.  But, I have to take my hat off to A Visual Introduction to Machine Learning.  Stephanie Yee and Tony Chu have a done a wonderful job with animation to show how decision trees - one common machine learning approach - actually work.


• The volume of data that IIoT threatens is going to challenge machine learning and Moore's law.  Simply put, there might be so much data that existing approaches to machine learning are overwhelmed.  GPU hardware is part of the answer, but This Startup Has A Brain-Inspired Chip For Machine Learning.  In other words, we might be moving towards specialized hardware in order to keep moving the start of the art forward.


• Is Big Data Too Risky?  OK, I will never tire of saying this...Repeat after me:  Big data does not equal Hadoop.  Too many (self-serving?) articles are written as if Hadoop is the only option, but this is a myth.  There are plenty of other ways to manage big data, including NoSQL databases and complex event processing engines.