Blogs

Hidden Gems at the Informs 2024 Business Analytics Conference – data structures.

By ken fordyce posted 05-01-2024 00:35

  

The yearly INFORMS Business Analytics Conference is always packed full of highlights from the keynote speaker to the Edelman Competition to theme sessions to presentation of the Wagner wining paper (this year Intel’s The Sort-Assemble-Blend Routing Problem and its Application to Semiconductors).  I always find hidden gems in the Vendor sessions on Sunday. Here I want to briefly highlight two of them.

·         Irv Lustig (Princeton Consultants): “Pandas for Analytics Practitioners, with Applications in Optimization

·         Frederic Gardi (Hexaly): a New Kind of Global Optimization Solver

Details about all of the sessions can be found at

https://meetings.informs.org/wordpress/analytics2024/agenda/exhibitor-workshops/

Briefly Irv stressed the importance of data constructs for successful optimization modeling – he introduced the term “tidy”.  He then provided an excellent and compelling overview of the use of Pandas to support “feeding” the beast.  I strongly recommend this to any venturing into Industrial Strength Optimization (ISO). I was reminded of an IBM modeling conference in Boulder, CO I attended with Gary Sullivan where Gene Woolsey gave his presentation: “Where is the data?”.  Irv’s slides are available from Princeton Consultants.  Irv writes often for their blog (https://princetonoptimization.com/blog/).  He is always a great read.

Questions that popped in my head while attending his presentation were:

1.       How do Pandas compare with SQL,

2.       For Tidy data how would the date effective data constructs used in the 2000 IBM Edelman Finalist application for central planning fit this construct

3.       How do Pandas compare to the general array constructs that were standard in APL2 by the early 1980s (APL2 at Glance by Jim Brown)

Fred provided an overview of the “Hexaly solver suite”. A key element of his presentation was the need to switch your mindset from using binaries to sets to handle classification and logic. For example, assume the firm makes red shirts or blue shirts on any given day, but it can only make one color each day. We are all wizzes at the binary constructs. It struck me that binaries are not the way “regular” folks would organize this knowledge.  They would do the following:

·         allowed_colors ß {red, blue}

·         the constraint would be

·         selected_color has length 1 and an element of allowed_colors

The use of sets to solve Sudoku can be find at

https://blog.arkieva.com/an-artificial-intelligence-based-solution-to-sudoku/

0 comments
3 views

Permalink