Do you have a second to talk about recommender systems? 🙂 a fun activity over the break

Edit from Jeremy: After chatting to @radek about this, we’ve decided to team up and do lesson walkthrus together – see this thread for details.

I have been working on a side project involving recommendations and I think it could be a fun opportunity to learn about:

  • How to set up an ML project
  • What is the most important component of an ML project
  • We could use what we learned in the lectures thus far to implement a very prominent paper and go through the steps of how to do so, what we already know is plenty enough
  • How to write fast Python code and why some code is slow
  • What are embeddings, really (and why they are the coolest thing ever :smile:)
  • Maybe we could even do some collaboration via github at some point
    … and probably other things as well :slight_smile: These are just some of the things I have encountered in the project so far, could plan some fun activities around them

Essentially, it would be taking what we are learning in the course to a new setting.

But I do not know if there would be interest?

I have a vision for what we could do, but in the end you can never know how well such a thing would work out or not :slight_smile:

Before the course started I had a vision I could maybe show people what homework might look like, how we probably all should be going through the lecture notebooks and engaging with the content (there is a way to doing homework!) but three things happened:

  • everyone seemed to be doing just fine judging by the energy and activity in the forums
  • Jeremy will capture all of that and much more in the deep dive from what it seems
  • doing such a thing would be a big step outside my comfort zone, it probably is easier for me to organize something smaller (plus, both in terms of time and emotional energy, probably my reservoirs of either are not at a level right now to easily accommodate such a major activity as creating recordings :slight_smile: )

But if you think that learning a bit about recsys might be something up your alley, and practicing what we learn in the course in a new setting, let me know :slight_smile:

In the least I can share some things via this thread with you and also open to walking through some things in a video call at some point if that might be helpful :slight_smile:

Let me know your thoughts please :slight_smile: Happy to do this if there is a chance it could be helpful to someone :slight_smile:


Hi radek, I am very interested in learning and making recommendation systems, I have a few projects in mind but don’t have the knowledge for it yet. What is the minimum level of knowledge required to be a part of this?

Being in the course :slight_smile:

I am not sure how much we might learn about recommendation systems, but there are things about ML that I think I could highlight through this project that can be valuable regardless of the domain :slight_smile:


I’m looking forward to this. Do you have a day & a time in mind for when you plan to do the video calls?

1 Like

@radek - I am definitely interested. I have never tried recommender systems outside movielens dataset. Would definitely love to learn along with you.

Adding sample datasets for recsys

Source: Datasets – RS_c


I like the idea to practice going from paper to code. Can you share a link?
I am dedicating most evenings (West Australian time) to the course, and on that basis I’d like to join in.

p.s. If its outside your comfort zone, you are growing and learning. So jump in.


OMG @radek that would be awesome. I love the fastai community and I love what you are doing by sharing and being so helpful and open, Radek. I understand that it is scary and outside of your comfort zone but please keep doing what you are doing – you are making a difference and your efforts are appreciated.

I know very little about recommendation engines - except what I learnt over coffee from a colleague who just left meteorology to go build them for a finance firm here in Australia - anything you do in this space would be amazing!

Sign me up.


Thanks a lot! I am very much interested in learning these topics with you and others!

Could you give us a slightly more concrete sense of starting line for this project, so that some of us who are beginners can be assured that we are not lack of something big before starting the project?

Yes, curating a project like this with people at different levels and paces can be challenging and exhausting, but I think all of us very much appreciate your willingness and effort of doing this!


Lovely, so there is some interest :slight_smile: Let’s do this!

There will be technical bits here and there, but what I would like to really do is share with you a little bit how I approach ML projects. There are no prerequisites for that :slight_smile: Essentially, how to apply what we are learning in the lectures in other contexts.

Everyone will be welcome to go as deep as they would like (there is no bottom to this endless pit, similarly how it is for nearly any other topic in ML :smile:), I will share material with you and maybe some suggestions on how you can go deeper, but the core will be literally applying what we did in the lecture, with the twist that we will use a dataset that is very popular in research.

I will provide everything anyone might need up to that point, and show step by step how we arrive at the basic set of results (assuming we will manage to implement the work in the paper, which is not certain :smile:). But I think there will be ample opportunities for people to take this further.

So no worries! That is the beauty of that it is a safe space to experiment!

If I mess up on being able to convey anything valuable to you, nothing bad will happen (though I will feel very bad about having wasted your time :slight_smile: )

If someone won’t be able to keep up because I mess up, or because they chose to binge watch Netflix instead of looking at a bit of code, again, nothing bad will happen! :slight_smile:

You cannot fail at :smile: If there was a chance to fail at I am quite certain I would have failed out very early on.

More info coming soon :slight_smile:


Awesome initiative Radek!. Definitely interested in this, and the video meetups.

1 Like

It would be wonderful to see a legendary alumni to show us how to approach ML projects in practice!

This is amazing! I am really looking forward to it! We all should do it as advised in the course and the book. But having your example done in front of us, not only prove it is doable, but also will be very inspiring to us all.

Another great work I am very much looking forward to!!

1 Like

Count me in, Radek. Great idea!

1 Like

After chatting to @radek about this, we’ve decided to team up and do lesson walkthrus together – see this thread for details.


Any recommeded materials to get started with recsys?

That is very meta :smile:

The upcoming lecture in the course will be a great place to start. It will cover a technique that is the best one to attempt at first on any recsys problem (right after building a baseline).

This lecture by Xavier Amatriain is just superb.

The blog post by my colleagues on the four stages of recommender systems.

I would then watch a video by Even Oldridge, here is one great to start with.

I don’t mean to promote myself, but if I am to give an honest answer, I would also follow myself on Twitter :joy: I post things like this and this there. If all goes well will also share a lot of code that I think can be quite helpful.

From there on, just going through other people’s code and writing a lot of your own code.


In addition to the stuff @radek mentioned, have a look at the collaborative filtering chapter of our book.


Super interested in recsys. I am joining the walk-thrus live but it would be great to have a couple of sessions on recsys separately and to keep this thread going.

1 Like

They will come, don’t worry! :slight_smile:


Great idea, I’ll be following this ! :raised_hands: