In 2021, I was fortunate enough to get the green light from my company, ShopRunner, to open-source Collie, a deep learning recommendations library I have been working on off-and-on since my internship two years ago. It has been a journey developing the library, but as I write this blog post now, we have officially A/B tested Collie against both our previously instated product-product and member-product recommendation algorithms and found Collie to match or significantly exceed them, respectively, in email CTOR. This means that, as of today, ShopRunner is officially using Collie as its primary recommendations system!
To celebrate these milestones, I put together a three-part blog post series on Medium walking through a simple use-case for Collie and slowly building up a worthy recommendations model. If that sounds boring, don’t fret – it’s about beer, and I end up trying the beers the model recommends to me!
Will I like it? Will I hate it? Read on to find out!
I would normally try to include the entirety of the blog post here on the site as well, but I haven’t found a good way to embed the Gist in this blog that looks pretty. The great news is that the blogs on Medium aren’t behind a paywall, are ADA compliant, and just generally look very pretty.
Fetching Better Beer Recommendations with Collie (Part 1)
Getting data, training a model, and talking about beer!
Fetching Better Beer Recommendations with Collie (Part 2)
Saving some time by training better models.
Fetching Better Beer Recommendations with Collie (Part 3)
Using metadata to train better models and (finally) drinking some beers!
All the code used in this blog post can be found here, and the entirety of the Collie library can be found here on GitHub, PyPI, Docs.