Data Science - Creative Collages, Zany Art & the Drunken Monk

Learn Data Science
Teradata Employee

We are discussing about Multi Genre Advanced Analytics (TM) in this post. Multi Genre Advanced Analytics (TM) is a postmodern development where seemingly subjective art & multi-faceted approach substitutes the past pursuit of rigid utopian explanation of  problems and solutions with a handful scalable techniques.

So what exactly is Multi Genre Advanced Analytics (TM) ?

Multi Genre Advanced Analytics (TM) is way to approach a business problem with a myriad of analytic approaches to express insights. The approach mixes Data Science and Art to bring the insight to the business and communicate that in many dimensions - obviously to increase the understanding of the problem and to pursue further discovery objectives. The key thing to this approach is the ease with which you can combine multiple approaches or algorithms seamlessly. In other words, we don't try to stereotype a problem with known problems and solutions learnt from school, instead go at it like a child with a creative approach keeping it scientific, measurable & repeatable ...

And so what's different about Multi Genre  ?

A few ideas characterize Multi Genre Advanced Analytics:

  • Big Data - no sampling and such. Sampling creates bias and doesn't lend itself to individual interactions and prevents personalized analytics. This is extremely important because there are some Genre of Analytics where sampling removes the signal that can loose signs of hacking or fraud.
  • More Behavioral, less Profile based - No assumptions on gender, age, demographic etc., which are rigid and habit forming and also not legal ! ...
  • No assumptions on structure of data or constraint on sources - ALL DATA OK!
  • Creativity - Breakdown of 'utopian solution' and no stereotyping problems. Every problem is unique and deserves the best solution there is available.
  • Fail super fast and try different hypothesis real quickly - a hypothesis that can span multiple genres!
  • Overloaded Analytics - Try as many analytic approaches under the sun if you can - no need to cherry pick anything ahead of time. Let data decide the algorithm ...
  • Lack of a rigid structure or assumption that most problems can be modeled from one or more best practice templates or play book- though ideas could be borrowed loosely from work before.
  • Awesome visualization to express the insights - Use insights as a business communication tool.
  • A quantitative backing to insights found so Dev Ops can pick it up w/o stopping at visuals
  • Data drives Analytic approaches not the other way.

"Multi Genre minimizes metanarratives ... "

What are some examples of Multi Genre Advanced Analytics (TM) ?

There is a churn problem. Traditional way to do it is 'logistic regression' even on big data. Logistic regression requires feature selection and analysts typically go through some variable elimination process such as dimension reduction to keep it tractable.Multi Genre Approach:

  • Yes, we can certainly do logistic regression, how about looking at event paths customer took to churn ?
  • How about looking at consecutive or co-occurring events and applying naive bayes to do a discriminate churn score calculation  ?
  • Should we try an ensemble technique to combine results from Naive Bayes/SVM or use Gradient Boosting for getting a better score ?
  • How about look at a customer cohort or connected graph to find the minimum distance to a churner and add that variable ? How about Page Rank or Clustering Coefficient in a social graph ?
  • How about trying XYZ that was discussed in a white paper ?
  • How about all of the above ?

While seemingly unstructured or unconventional way of doing analytics, multi-genre analytic approaches can unravel hidden structures in your data in an explainable way and as the analyst wishes to  approach it. Not only that, it's effective and more precise and personalized than traditional approach. This is primarily due to variety that exposes disconnects in understanding of the problem!Other examples include system part failure using a combination of Symbolic Aggregate Approximation and Text Analytics, Finding non-compliant users using Graph, Text, Trade patterns and Email meta data. The list is endless. KDD & Kaggle Competitions also use ensemble techniques as opposed to picking one winning algorithm which  are one of the cool examples of the brave new approach.

What are the challenges with Multi Genre Advanced Analytics (TM) ?

A talented few, versatile with command line and deep algorithms can do it in any platform - kind of. For the rest of us who want to keep an eye on the forest, there are a couple of requirements:

  • Seamless platform that allows iterative discovery and combining multiple analytics including Visualization
  • Equally seamless way to measure accuracy and repeat it and move to operationalize and create powerful solutions ...
  • Not just restricted to folks who understand algorithms in depth, but allows experimentation w/o loosing sight of correctness in results.

This is exactly where platforms like Teradata Aster can help in closing the gaps. Check out for more ...

Also link to Teradata's Art of Analytics gallery that demonstrates the solutions for varied problems using a multi genre approach - algorithms & visualizations using the Teradata Aster technology - emphasis is on visualizations.

Sidebar on postmodernism definitions:

Simplest example is iPhone/Android vs an Old Cell phone. Every phone is individualized and personalized with a smart phone. Every user decides the experience in some way ... With the old regular phone, everyone's usage is limited because of the way the solution is offered to them.

For further edification, I'd google 'postmodernism'  - there is plenty of other examples ..