Matrix Factorization Useful to Whiskey Production

Who ever thought that something you learned in class would actually be useful to you on a Saturday night when you wanted to just relax and kick back with some friends and your favorite bottle of scotch whiskey?  Well, the authors of this article, “Clustering of Scotch Whiskeys using Non-Negative Matrix Factorizations” from the SPES/Q&P Newsletter, applied uses of sceintific computing to the world of alcohol production. 

 These authors were presented the task to cluster the data set of 86 scotch whiskeys rated on a 5 point scale for 12 flavor characteristics.  Pure water and grain nuetral alcohol have no flavor.  Therefore, all these flavors must be engineered into the whiskey by the still master through fermentation of secondary compounds and different aging processes.  Now is there a way to keep all your whiskey drinking friends happy with their particular tastes using relatively few of the 86 brands?

This problem is ideally suited to perform non-negative matrix factorization (NMF).  This method produces 2 matrices, W and P, and an additional error matrix. 

X = WP + E

The matrix X is positive, consisting of the 86 whiskeys rated (0 to 4)on the 12 flavor characteristics.  W and P typically represent a weight matrix and a profile matrix respectively.  Therefore, by factoring the rating matrix of the whiskeys, the P matrix would represent the prototypical flavor patterns and the W matrix would represent the mixing levels of these flavors.   

Since the 12 dimensional world of the whiskey flavors is rather large, the authors simplified this problem down to four flavor factors.  These four flavor factors were a combination of the orignal 12, one having more of the sweet and floral characterisitcs, another having more honey and malty characteristics, etc. 

After factoring this matrix, the authors found 8 whiskeys (4 for each flavor factor) that had extremely close flavor profiles to the individual flavor factors described above.  They continue to propose that stocking only these 8 whiskeys may be a cost-effective way of keeping all your whiskey drinking friends happy by mixing these base flavored whiskeys to match your friend’s taste.

http://www.niss.org/irMF/ScotchWhisky.pdf

Posted in Topics: Uncategorized

Jump down to leave a comment.

2 Responses to “Matrix Factorization Useful to Whiskey Production”

  1. Боевые действия во Львове. Женщины отражали атаку на одно из крупнейш Says:

    […] Matrix Factorization Useful to Whiskey Production […]

  2. Скандал вокруг фильма о чернокожем внуке Юрия Гагарина. Says:

    […] Matrix Factorization Useful to Whiskey Production […]

Leave a Comment

You must be logged in to post a comment.



* You can follow any responses to this entry through the RSS 2.0 feed.