GitLab Bookclub - Hands On Machine Learning, 3 Mar 2022

Previous Meeting

⏯

youtube image

►

From YouTube: BookClub - Hands On Machine Learning [EMEA KICKOFF]

Description

BookClub Issue: https://gitlab.com/gitlab-com/book-clubs/-/issues/40
Discussions Repository: https://gitlab.com/gitlab-org/bookclub-hands-on-machine-learning

A

Okay, so everyone, hello everyone, welcome to this kickoff of uh this book club. It's the first time. I try a book club here in gitlab. uh I know how this is gonna go and I will try to be as much async as possible on this. So I created the keep the agenda over there um and okay. So I don't know everybody has the book already right or close to getting? I got mine over here. It's a very sizable book, and this goes into my first point.

A

I don't think we can. We can expect ourselves to finish this entire thing. uh It's a lot of stuff. It's uh I don't know and at some point, uh if we try to do it, people just report and go out come out and then expectations are broke. I would rather start with low expectations rather than high expectations than this and find something we can now be proud of accomplishing by the end of this, of the time that we're gonna have this. No, that is a book club.

A

uh Oh yeah, again everybody can talk. uh I I'm not keeping a talk over here or having a conversation so uh going back over here.

A

Yeah, so I also forget to mention that the the first part is really cool, it's more about machine learning in itself or in a more simpler term, but the second part is really neural networks, which is also really cool, which is also really have its own, uh very much uh application very strong applications, uh but it's also a little bit of invest, especially how it goes through on this book a little bit of an advanced topic, but it's quite fun. So I don't know uh your thoughts on this.

A

What what would you like to uh to take out of this uh of this uh this time? We're gonna, be studying this book.

B

Yeah I can, I can start so, um like I wrote in the agenda, I'm I'm pretty bad at finishing books, um I'm good at buying them, but uh actually starting to read them and then go through them is a whole different story.

B

So yeah, um I'm just hoping that this will keep me like accountable to keep going on this and get some consistency in going through the book and also like ml has been a thing. I've been wanting to look into for a very long time, but never really take the the time out of my day or or anywhere to really get to know it at all so yeah. I already already feel the first chapter uh has.

C

B

Taught me a lot yeah.

A

Yeah, so I I I don't think I'll be finishing this book as well. I don't think I'll be reading all chapters. I I read I'd like to skim through it and I select a few chapter chapters that interested me they're more interesting to me and then I'm gonna be focused on them. um Otherwise it gets really boring. You should try to force yourself into the book. I mean it can be rewarding by the end. But for me as a person I like personally, I don't don't really like that, and the.

D

D

So yeah, I hi everyone by the way and I definitely support some uh and I think we are the same boat. So I'm good at buying technical books or technical courses going halfway through or to some extent and just then switching to something else.

D

But I tried to start with machining learning and the uh I did the domain around it a couple of times uh getting in touch with some bases uh like linear regression, and it's this very, very beginner uh approaches, but never really get down to the uh actual neural networks. uh By looking just at the uh prologue to the uh to the book, uh I thought that maybe it would be cool to like skip related between these two sections. So touch start. Probably.

D

I think it would be very beneficial to start with the uh with the intro, because they they want to explain the the fundamental concepts. This that's what the author status and I think at least from my perspective.

D

It would be nice to to have my like found them found dominance like debate the baseline well set and then uh go somewhere else, because in the latter part of the first section of the book, they say like they're, going for the fine tuning and the like uh into the details, how to deal with the machine learning without actual neural networks.

D

And maybe this is a place where we could skip and say, like okay, we know the basics of the machine learning and let's go to the neural networks to at least see this and get like a touch or a hint. How this may look like because well it looks extremely interesting, but probably maybe hard to get into without getting at least basic information from the machine learning part.

E

The next one was me, but I think we skipped ethan. You wrote before us, so if you want to you can go now. I.

A

Was not following, I guess I was just like okay, everybody wants to talk about something.

F

Yeah, I thought it was a free-for-all. So sorry, uh um so it's basically like sam said I do the same thing: buy books start reading them and then they pile up and I feel good for having them and feel accomplished for having bought them. And then I move on to something else. So I uh and I've done part of andrewing's uh coursera course.

F

I've done some other like machine learning courses, but never like a full course of study, uh never a like a full tech book, it's more like the uh the tutorials and the the the little youtube things.

F

um So I'm looking for kind of a more structured um learning environment and I I'm because I've done some of the other stuff like I I can understand quite a bit of the first part and the the second part is really interesting as well, uh but I'm fine with doing either uh jumping around I'm pretty open, because I'm just here to learn what I can um in a in a way that encourages me to get to the end.

E

Cool cool, so the next one is me. My name is right: I'm working at gitlab as a data engineer and I'm studying a master's of sciences in artificial intelligence at the yotkawa university in linz, and I think the part I think I actually know the part that interests me. The most is ml ops. So whoever builds a machine learning algorithm, pretty cool nice at the to that moment and on. We need to actually serve it to somebody, and I really want to learn how to do this from start to end.

E

It would be cool if I can implement a model myself, but you know you can the very easy models you can just read through an article on the internet implement one and then it's done. This is just the basis for me. So it's the input. I would really like to do. Ci cd for the machine learning models, debug them problems that can come retraining stuff like this. So that's why I said the deep learning part.

E

The second part of the book is not the most interesting for me right now, although I do have to pass some exams at university for it, but I will deal with it separately, so yeah. I would love to join the parts where we discuss, maybe the first part of the book I also had shortly. I wanted to mention that it's totally fine for all of us to only read the technical book on the parts that we care about. I mean this is not the novel.

E

The trick, the trick with our book club is that exactly this is not a novel, so we need to either agree on the chapters we want to check out or we can separate in groups. I guess.

A

We are 20 people right, there's 20 people uh like are not on this one meeting specifically, but there were 20 people who added the blue book or on the issue. So there's a lot of people will be interested on each specific chapter. uh We can break down later by chapter by group by go by background, whatever time zone people who want to do their reading, rather than always or all folks are implementing their workflows. This is a very applied book so that it has code for everything which is great.

A

I guess for for all of us. uh This is uh what I enjoyed I enjoyed about this um yeah. This is the first time I run a book club this large and I am learning as I go and I don't want to see be the person running it as well like I don't want to be the one I I want this to be more of it.

A

I would like this to be more of like a self-guided thing by each one of us but yeah like I said it's not a novel, uh it doesn't really matter keep up if we skip one chapter or not, it's actually really cool, because if each one, one of us is really different different different chapters, we can create summaries for the others of what is important or what is who are the main learnings and then it's just easier to to go through everything right or record small video or do something which is also great for people who come afterwards and want to learn more about everything that we're doing here.

A

So it's a good format. I guess.

A

We have roy and then we have alexander.

C

Yeah sorry um yeah, I'm not very familiar with machine learning right now, but I always find it interesting. So at this point, I'm just joining the book club to get more familiar with it and yeah get some motivation to actually dive deep in it and actually finish the book like most of us here. I guess so. Yeah I'm open to pretty much learning a little bit about everything or typing deep in deep in one topic, either is fine for me.

A

And then you have finally alexander.

G

Yeah so hello, everyone and it's for me. It's a new experience so never participated in such book clubs and also I'm well, I'm interested in machine learning and in two months it will be nice to just get an overview um principle understanding how it works, and for me it will be special interesting how I can apply it in a in real life. What kind of tasks and practice maybe visit a little bit so after that to make it possible to just dive deeper.

G

If I want to and to spend more time on it, so it will be great to just play this probably some code and do something actually, so that's that's it and the whole course sounds promising. So thank you for creating that uh direction, and I just I was happy to to see it and to join it.

A

It's a fun thing to do: it's also a really great way of connecting with people across teams, um because we're mostly like on bubbles within our own team, especially on a remote setup that we just do work with the people around us, except for code reviews that we have a little bit more exposure to other people. But outside of that, it's really hard to get the opportunity to interact with people. And I've learned that from these interactions, a lot of new ideas can.

C

A

Up um when you start discussing the use cases of every like uh different verify code reveal and then other stages. You get ideas uh coming up and from ideas. You get really nice cool products, really nice cool features uh or uh cool collabs between different departments. It was.

C

I always had a lot.

A

Of fun with those um okay, so I had some points on the document.

A

Yeah, so the first one we already spoke about a bit is about setting the expectations. um Like I said it's it's going to be hard for us to go through the whole book. I did. I also mentioned about cincinnati novel, so I'm thinking about how can we be proud of not proud, but like okay? This is really cool that we did uh over this five months and not feel like.

A

Okay, I lost uh one of us like feel like they lost their time within this book club um and one idea is just choose uh going through regards each one: just chooses four or five uh chapters that really pick up our interest uh and then go from there. So, instead of okay, I'm not I'm gonna read the entire book. Now, just let's focus on a few uh chapters.

A

I think that chapter one for the people who are not coming here with uh with uh machine learning, so at least chapter one and two um one, I would say one through four or if you don't know, machine learning are good um they're more about the concepts behind machine learning and the the ideas everything, but the following ones are more of about algorithms or techniques, uh chapters from five until nine and those are optional. So we don't need everyone to study study. All of this all these chapters might be that some people are interested.

A

For example, I am more interested in the ensemble learning and the support vector machines, but I will not be reading the first. I will just discuss through the first four chapters and but on the other side on the part, two there's also some chapters that I want to look at, not all of them, but just one or two.

A

I think if we do that, it's a lot easier to accomplish to feel accomplished over this uh uh of over this time. I don't want us to go through a full quarter or like six months, trying to finish this and everybody's just pushing it, and then it's just like an annoying experience.

A

um So I don't know it's one idea or I don't know how. So if anybody has other ideas and how to approach this, I have to coming up with a use case and finally, through the book, how to approach that use case. I don't know as well.

A

E

Yes, I like this. I I like this idea very much. I actually have already underlined my chapter, so I was prepared.

A

Okay, good, so just to get an idea out of the folks that over here who never had any exposure to machine learning like I never had any study or machine learning. So far, when alexander did you had, uh did you study a bit machine learning before making some no.

D

A

D

Tried the basics like the inner regression or knife bias classifier a little bit of pandas uh these kind of things, but nothing very advanced.

A

Okay, now it's just more of uh for your st rather than anything. I think this book is exactly for people.

A

It has a lot for people who never did machine learning uh the first two chapters are really good for people who never machine learning, but it also has, for example, that I really like the swim chapter, because it's very thorough for people who even know and have experience with machine learning already.

A

So exactly also, if you hear the chapters itself, if it's getting too complicated to get into complex- or it might be that within the chapter there's stuff- that just doesn't click or it's not really important, uh it's fine for skip. No there's! No book police, there's gonna, be stopping your house after and say: okay, no, you didn't read that specific chapter or whatever it's it's energy. You do whatever you want. It's a book club all right and yeah.

A

So definitely one to four are really good for uh beginners and- and that's the second point that I had over here. uh Well, it's part of the second point. I already created this leg channel for four updates, but one thing is: how do we go about discussing things?

A

We have the issue where I'll be posting recordings and of the syncs, but if we want to discuss a a problem or a question or whatever we can do that in slack, but the problems that select eyes uh the message on select the eyes over time and then the next folks that come out want to have the club that won't have that um it could be on the google docs, but I think especially for this type of discussion, that's more technical code and stuff like that. It's not really made for it.

A

I don't know ideas.

E

um What is this second comment here? Eduardo and sam wrote that having repository where questions are added and everyone can see and respond, and that's some sort of of notification to the slack channel that a new issue was created. Does this mean that we create separate issues per question? That's a bit too much or.

A

Yeah, I don't know, I I think it's a bit too much. If I have a question for an issue, it would almost work like a stack of her 123.

A

It can be an issue per chapter, perhaps where we have.

E

Actually, this makes a lot of sense now, more words say more words.

A

E

About this idea, with the chapters.

A

Yeah, so you have one issue per chapter of the book or per chapter that you're gonna read and we tag the people that are gonna read the chapter right like the people that chose the chapter to be reading and over there. You have the questions. uh Like you add one question and the discussions, so each question becomes one point on the thread and you go about that. The good thing about having a repository is that we can create branches with code um for the questions itself.

A

So if you want to implement stuff one that would cluster, we can even add a call that we're trying to work on or whatnot and have the questions on quoting itself. I don't know it was one idea that I had I it was I I actually checked some over there on that comment and I think I'll just remove this okay yeah.

A

A

Perhaps perhaps one issue per question is too much more, but one issue for chapter and questions and then, if one question becomes too big, we can move that. We can promote that question into an issue or something that will make our sense.

E

We can all be like notified so either actively. We can go to these chapters. For example, if I only care about four chapters, I will set myself to get notified only on these four chapters and whoever comments there. I will get these comments. This is.

A

Yeah, nice think it's a good type of I like that one I like that idea uh and then, in the end we can summarize learnings on the issue description.

A

So for the people who read the chapter, they can create a summary or a small, video or whatnot, or I don't know a musical about the chapter itself. If that's what they really like, but well some some summary of the learnings. It can be like a sunset. Each one requires a loop uh of quick learnings or just a a post or whatnot free for all so yeah. I guess.

E

A

Israel uh repository it's gonna, be one repository right. We just need one repository, that is a book club repository project or let me see this issue the the issue of the book club, where I don't even know where it is. I just clicked on a link where I had a okay, create issue here and I don't even know what repository is that.

E

A

E

A project but it's empty, so you can have the honor of doing the first commit.

A

Okay, now it's uh yeah, it's just a project. uh I don't have any. So, let's.

A

Okay, uh will someone do the honors of creating uh that for us creating a project and within the book club, if you don't have permissions, I think uh samantha lee can can help with that.

A

So she owns that thing. Okay, there is oh.

C

There's another booklet.

A

Because there's also this book, there's also this over here- uh there's also this that it's all book clubs which, uh where the issue that we have for this club exists so but there's also a repository for okay, okay,.

A

Who created this okay? So anyway? Oh there's, an issue, hey! Oh! This was five years ago. Okay, so.

E

We can, we can call ours, a technical book club.

A

Okay, so who who takes the lead of creating this? Who goes now? Who who volunteers for this? Just setting up the repository? We will figure out how to write the repository later.

A

I need a volunteer or a volunteer someone.

B

I've kind of lost track so we're trying to create a.

C

B

For this this book club, but it has to be under wet.

A

B

A

Be under anything really it can be a connected laptop or it's just that there is this project for clubs or we have an issue for this book, but I think it should be another one. Otherwise it just mix everything over there, perhaps just a new repository for this- uh a new project for this uh another repository value project for this book club, where we can start adding the issues uh for each chapter.

A

So if you choose your chapter uh and you don't see an issue for that chapter- create the chapter, create an issue for the chapter chapter. Otherwise just leave your classes, yeah, otherwise, join the tag yourself on the chapter for that on the issue for that chapter. So let me write it down. So let me I'll go over here.

A

A

Create a project and once we choose the chapters.

A

A

Create issues for each chapter: if they don't exist, they don't exist for beginners.

A

uh Oh, let me add it over. There.

A

For beginners chapters 24 are recommended.

A

Yeah, so I still need a volunteer.

D

I'm creating project as we speak. Okay.

B

A

All are okay, so yeah cool here.

D

A

Goes good good, not like a little bit of pressure? Okay, so we have this repository and then we create issues as we go.

B

And see why my slug was taken now.

A

So cool that's solved.

A

And I think this also goes over some of the next question right ideas on how to make this as much strength as possible. I think with having the the everything through the issue uh through the the project that we create. uh We can help a lot. Of course. We're gonna have some things here and there uh it's nice to have things and talk about this, but I don't. I wouldn't like this to be the biggest uh biggest driver for these discussions, that we have it's nice, that other people will have access to this later.

A

C

A

On how to make this, uh how to improve and sync experience for this.

D

Just a wild idea, uh since we are working for the issues for each chapter, while we like everyone who is interested in the issue or in the chapter once we feel like okay, we exhausted the topic. We reached the uh satisfying the outcome. We then make could set up the sync call and, like summarize, what we learn- and this would be the summary of the um of the this chapter and like we put it on the top of the issue in description and anybody who is coming back uh in the future.

D

When the book club is finished, they could just go over and see uh this summary like ama or like ama free-for-all, something like that. um Just just world idea.

D

That's where we, I guess we're getting the both assing first we're starting with reasync and the summary of some uh kind of getting together and building this connection as well.

D

I like that idea.

A

I also like this idea quite well actually.

A

And it also allows people to go through the book sometime in in different ways right, because some people are going to be more interested on the on the on the codes. People are going to be more interested in the concepts or someone might even have a specific use case that they want to explore.

A

Okay, I think it's a good way, good idea, any other ideas.

F

I think something like that and coupled with an issue per chapter, to capture the uh all the questions and the back and forth and everything that's and then using slack for notifications. That's going to go a long way to making it async um and that I think, that's a good place to start and we can kind of iterate from there and see what see how we can tweak it.

B

I created an issue for the first five chapters.

B

And I think it would also be quite cool if we, if we want to share code snippets, do like a folder per chapter and then everyone can uh just create a subfolder for their own username to push code to there.

B

A

B

F

I was thinking, branches might be, might be a good.

A

Now that we have notebook tips, we can even have like almost called reviews on the on the call that we create for this and welcome.

A

But it's uh yeah, I think it's a good idea.

E

I was thinking.

A

E

That, while we walk through the chapters, we might need to write the code ourselves, but I am so sure it has to be somewhere on some git repository.

A

Yeah this is this is available online. uh The code for this uh this book is available online. It will be more if you want to try out uh variations. There are exercises by the end of each chapter that we can try to implement as well. The answers are in the back of the book, but there are some exercises that are somewhat interesting. If you want to get your hands dirty, um so we could discuss the exercises as well um for the chapter or something like that, we can even precede the issues of.

A

I think that's a good idea. Actually, I'm gonna appreciate the the issues of the chapters with the exercise, questions of the chapter chapter in itself, so I'm gonna add one question for each chapter. Then uh it's already a way of discussing uh I'm.

B

A

Even write it down.

A

A

The issues repository the issues, chapter issues.

A

With the questions exercise, questions.

A

Well, there's not something else like as a precedent system.

A

Precedence is just weird. It's unnecessary, okay, good, oh yeah, uh thanks! I'm gonna add this into the main issue of 40.

A

uh I added that into there. Let me make a proposal.

A

And I'm gonna write down this to the protocol that we have on the main issue. uh For this thing uh right, that's the decisions that we made here. um um So the schedule will be pretty much at least from my guide here. The schedule of next uh acceptance keep up we're gonna have one another one next week, so it's gonna be very much uh ad hoc meetings, since I want the people that are comfortable in the chapter are okay with.

A

uh When I have a sink or they are okay, we can add a an emoji uh so that they know that everybody is done with that or that chapter or is going through, or I don't know on the chapter some people might not want to, and it's okay, uh if you feel like this, is not delivering value.

C

A

You or not, that the value you would expect it's okay to drop. The worst thing is to lose your own time over this there's, no pressure at all. It's not gonna show up on your performance evaluation, so I'm not gonna tell your manager or anything uh perhaps an entire month, but you know your your parents are or something. But you know it's uh it's okay. So if this is too much uh things come up, it's a parent is change.

A

We are trying our best here to learn something new. Sometimes it's not possible. Sometimes it's okay, don't feel uh pressured into continuing just because peer pressure well.

C

A

People here are looking exactly for that right. So if that's the case also cool uh how much you judge anyone, um so that's cool! Okay, um oh brett, just joined us. uh We had some discussions brought either on the on the document itself, uh just summarizing everything that we spoke about uh today. It was mostly about the uh the process that we're gonna follow. So each one. Instead of trying to read the entire book, we're gonna each one choose five, a few chapters like five or six chapters that we want to look at before.

A

I begin your recommendations of chapters one through four, at least um then for these chapters. We go into the repository that was created the project and each chapter is going to have an issue tag yourself uh into that issue, because that means like people will be talking more about that specific chapter over there. So, instead of having like a book club you're going to have multiple chat requests, you can think about it this way uh and then once everybody's ready, we leave that chapter.

A

We're gonna have a like they get together and get summarize their learnings and get their discussions. So it's gonna be mostly async book club and, if you feel like you want someone to nudge you uh and ask if you uh one are following through I'll, we'll ask and ask appear to help you out.

A

um If anyone else has any other comments or anything, you want to talk about already.

A

A

Good so I'll be uh stop recording and I'll be close. This off I'm going to give everyone else back, and thanks all for joining, I'm very excited about this. I hope it is as fun to you as it is to me we'll chat over the issues and overstep have a good one.

D

Thank you thank.

E

You bye thank you.

D