In today’s data-driven world, GPUs are the hardware of choice for training Deep Learning models. What about tasks that do not involve artificial neural networks? For instance, is there a benefit to using a GPU for making product recommendations? Continue reading to find out!

This article was first published on June 17, 2020 on Scaleway’s official blog and is reposted here for your convenience.

Anyone selling anything these days makes recommendations. “Customers who bought this item also bought these ones.” “Here are the top 10 TV series that we bet you’ll enjoy.” Sometimes these recommendations are based on the intrinsic properties of the products, but more often they come from the behaviours of users such as yourself.

Let us say we want to build a simple book recommender system. The data that we need for it is available on any website containing users’ reviews of books: e.g. this dataset has been collected from BookCrossing.com, a website dedicated to the practice of “releasing books into the wild” — leaving them in public places to be picked up and read by other members of the community. There are three data tables available, but we will only be needing two of them today: BX-Books and BX-Book-Ratings containing information on the books and the bookcrossers’ book ratings respectively (pardon the excessive use of book in the preceding sentence, finding a suitable synonym is no easy task!). Each book in BX-Books is identified by a unique ISBN, and each row of BX-Book-Ratings lists the ISBN of the title that the user’s rating refers to.

#data-science #pytorch #machine-learning #artificial-intelligence