Grokking Machine Learning

Hi Kashan Ahmed !
No, the book only covers supervised learning. Perhaps a future book will cover unsipervised

I mention some things about unsupervised learning at the beginning, but never cover the algorithms in detail

Thank you for answering. 🙂

Hi Alper Demirel , thank you for your interest!
I wrote this book to be understandable by readers without an extensive background in math and programming, so the algorithms are explained in a more conceptual way with examples, figures, and stories, as opposed to formulas. (The formulas are there too, but they appear after the conceptual explanations)

Thanks a lot for the answer, I’m so glad it’s not a book full of formulas. I hope I get the chance to read it!

Hi Wendy Mak !
The grokking books tend to make a big effort in making the topic understandable for most people, not just those with a super technical background. I always like to explain things in that manner, as that is the way I like to understand things. When I started writing this book it wasn’t meant to be a grokking book, but when the editors saw it had that style, they asked me if I wanted to make it part of the series and I thought it was a good idea.

Riffing off this thread, how did you go about identifying (a) that you wanted to write this book and (b) connect with Manning for publishing?

Wendy Mak I have always loved teaching. One reason is that I am a slow learner, so I always have to digest everything I learn until I make it very understandable for myself (with figures, examples, etc), and so when I explain it to others it has already been digested in my mind, so teaching comes out easier.
I started as a mathematician, my PhD was in math and I was teaching and working on research. The ML bug bit me when I started hearing about it and I decided to switch careers and started working at google, that’s when I started learning ML more seriously.

http://www.catb.org/jargon/html/G/grok.html 🙂

Hi Neal Lathia ! To grok is to understand something very well in a conceptual and intuitive manner. This is why the grokking books try to explain things in such a way for most people to understand even if they don’t have a technical background.

Thanks Alexander Seifert !

Hi Eric Sims !
Great question! I like many algorithms, I gotta say I really enjoy the kernel method (SVMs) because it’s so clever. Also naive Bayes, and anything that is purely probabilistic, since all they do is playing with conditional probabilities and they can do wonders.

Eric Sims , yes, that’s a huge part of the book, and the reason is that I believe that to start in ML, one doesn’t need that much understanding of math. The math can be learned as you learn ML, hand in hand. Most experts recommend learning a lot of math and then starting ML, which I don’t fully agree with, which is why I wrote this book.
As for essential math to read this book, high school math is enough. Having an intuition of what a formula is, basic equations like the equation of a line, and basic probability (e.g., being able to calculate a probability as a ratio of numbers) is enough, as everything else can be developed as we go.

Hi Chetna ! Great question, no, the book doesn’t talk about system design. It talks about the algorithms, how they work, and how to apply them. There is code, but mostly to study datasets, not to put in production.

got it, thanks 🙂

Hi Kashan Ahmed! Great question, yes, the person who finishes the book will have a working knowledge of most supervised learning algorithms, and the packages to use them (including a deep learning package), so they’re definitely an intermediate level ML practitioner.

More can always be learned after the book, including other similar libraries, other fields such as unsupervised learning, generative learning, and reinforcement learning. Also, ML production and system design is a field people can learn more about outside of the book, in case they want to implement models in production.

Thank you for detailed answer.

Thanks Krzysztof Ograbek!
The ML libraries are scikit learn, Turi create, Keras (Tensorflow), and xgboost. For other stuff, matplotlib, pandas, numpy

Hi xnot!
I think geometric intuition is more important than the formulas and the math. And yes, I think this can be done while learning and developing ML.
When I talk about geometric intuition, I mean things like drawing lines that pass close to a group of points, lines and planes that separate points of different colors, rotations, transformations, etc. Most people have this type of intuition, it’s only a matter of tying it to the algorithms and the applications.

Thanks Luis Serrano. Does the book cover ways of approaching this?

Thanks Quynh Le!
If you learned regression, then you know machine learning. Most of the main algorithms are similar to that one, only with some small tweaks. I think a book or introductory course in ML can get you up to speed. And yes, one can implement everything only in Python, as a matter of fact, it’s the one I recommend the most, since most of the important packages are written there.

Do you suggest any intro ML course or book (other than Andrew Ng course)? Can I read and implement projects from your book using Python?

I have a bunch of videos here that you may enjoy: https://serrano.academy

Also, there are some interesting courses at Udacity. Here is one on deep learning that I teach with a few other people:
Intro to Deep Learning with PyTorch

Quynh Le ^^

Luis Serrano Thanks for the suggestions, I am a big fan of Udacity courses! I’ll check your website as well!

Hi WingCode, great questions!

1. It is challenging to strike the right balance between oversimplifying while explaining, this is why I try my explanations on people, including experts and non-experts. If I can get a non-expert to understand it while not boring an expert, I feel that I’ve hit the right spot.
   
2. Yes absolutely, the book can be used to communicate data science to stakeholders, since it has lots of real-life applications, which they can use to see the value of these algorithms. It also explains the details, which can show them they’re not that complicated or hard to implement, and that they’re not voodoo, just simple math used with the purpose of solving the problem.
   
3. Ah interesting. There are many topics I’ve found hard to ‘grok’. In the book, some of the ensemble methods such as xgboost or gradient boosted trees took me a long time to understand and to simplify, since they have lots of ins and outs, and some of them sound a bit arbitrary. Once you really get into the details, they are not arbitrary, but it takes a while to realize. Other topics outside of the book scope, such as reinforcement learning, take me quite a long time, for the same reason. But I have a believe that everything can be ‘grokked’, and the ones that haven’t yet, is because we don’t fully understand them. 🙂

Thank you Luis for the great answers! :)

WingCode great questions, keep them coming! 🙂

1. Unsupervised learning is a big one, since it covers generative learning. Aside from that, reinforcement learning is one.
   
2. For this book, I picked the most popular algorithms of supervised learning, so it was pretty straightforward. As for other type of content creation (videos, etc), I normally have a list of things that I’m interested on because I’m trying to understand them, in topics like probability, ML, statistics, etc., so as I understand them, I create content about them. Some of them come out of projects at work, and others from things I watch or read, etc. It’s pretty random. 🙂
   
3. YES!!! Quantum is definitely next on the list. Right now I’m working on quantum ML, and learning a lot of stuff. My goal is to understand it in a simple way, just like it is with ML. Definitely keep an eye for that material, because I’ll be grokking quantum a lot in the near future.

Haha, thanks Luis. It is a pleasure to ask you questions & also to get your answers :)

Gracias Alex !
The book aims to be a one-size-fits-all, as it offers beginners a chance to get into ML, and the more experts a different and simpler view of the algorithms that perhaps they haven’t seen before.
But in a nutshell, the reader who’ll enjoy this book the most is the beginner in ML who comes without a very heavy knowledge of mathematics and programming. The reason is because in the book we explain the methods and algorithms from a more intuitive perspective, where the math is there, but more in drawings and examples than in formulas.

Hi Lavanya M K!
For the illustrations I used mostly keynote. For some of the plots and graphs I used matplotlib.

I like “Rules of machine learning” from folks at Google. I find it somewhat similar to ML Yearning

Thank you Andrea Mordenti, glad you enjoy the explanations, and thank you for considering the book for the course!
The printed version will be available soon, hopefully in a month or two, I’ll keep you posted.

Hi luckylittle, thank you for your interest in the book and the page! I hope you had a chance to check out the youtube videos.

1. For anomaly detection, such as Fukushima, I would use unsupervised learning techniques, such as clustering, etc. Some of them, like DBSCAN, have a way to detect outliers.
   
2. I add examples of recommendation engines such as youtube in several places in the book. For example, you can use linear regression to predict how long a video will be watched by a user based on how long that user has watched other videos. Also, classification methods can be used similarly to predict if the video will get watched or not. Also, there is an example of analyzing the text in Netflix reviews of movies, using python.
   
3. The book will be released very soon, it’s in the last stage of production. I think there’ll be a physical version in about 1-2 months. I have enjoyed the writing (and re-writing) process, as it has taught me to understand things in a different way.

Luis Serrano Yes, I have checked out your YouTube videos - very impressive!

Krzysztof Ograbek I recommend grokking deep learning, grokking deep reinforcement learning and grokking algorithms. All very good. DL and DRL are written by two friends and former coworkers of mine (Andrew trash and Miguel morales), who are great at explaining
I also recommend advanced algorithms and data structures by Marcella la Rocca, also from manning, great for the fundamentals

Thanks Matthew Emerick !
No, the book focuses on supervised learning, and there is no graph machine learning
The books I recommend after this would be grokking deepnleanring and grokking deep reinforcement learning. They are a good continuation of the topic, and written in a similar style

And as a follow-up, how did your day back then look like?

Thanks Alexey Grigorev!
Being at udacity was great because I really enjoy teaching and that’s what I was doing. Being able to lead a team that produces educational content was great because I would design the courses and was able to propagate my style to others, who would also inject their style into the courses, getting the best of both worlds. The transition to Apple was smooth since at Apple I was teaching workshops, it was actually fun because they were in person, and I enjoy that.
Coming from ML engineer to educator was very rewarding, because I think as a teacher more than as an engineer. The engineering mindset is more of a “get things done” mindset, while I like to understand things slowly and thoroughly, which is more suitable for teaching.

My day back then had a lot of meetings with instructors in my team, I would check their material and give them feedback. There were team meetings to decide on curriculum etc. I would try to crunch the meetings in some days, and leave some days empty so that i could sit down and create educational content for hours, as this can be a lengthy process.

Thanks for sharing it! Sounds quite fun