Hi Ashish Lalchandani! We are really glad being here presenting the Kaggle Book as book of the week!
Breaking the ice is one of the difficult things with Kaggle. I suggest you first should try participating in introductory competitions such the Titanic one (https://www.kaggle.com/competitions?hostSegmentIdFilter=5) or, since they are more lively and bit more challenging to Tabular playground competitions (https://www.kaggle.com/competitions?hostSegmentIdFilter=8). By doing that you can get rapidly a working knowledge of how Kaggle works and find your best way to enjoy about Kaggle and that’s not necessarily by competitions, you can participate in many ways, by writing your code doing eda or modelling, by studying others’ notebooks, by discussing on the forums about the possible solutions to the challenge, It is all up to you what you find more useful for learning.

Thank you very much Luca! Thanks for providing the links too, much appreciated! I usually practice on jupyter notebook/google colab, and am really excited to compete on Kaggle, but never knew how. Thanks for providing your valuable insights! And all the very best for your book!

Too much gamification sometimes leads participants to find shortcuts: voting rings, leaks, reverse engineering, various tricks that are far away for the real world practices.

Surely the learning part and the opportunity to try anything that comes to my mind in order to solve the proposed problem :-)

When it comes to tabular data problems, ensembles are still the best solution in most cases. In the majority of competitions, which are deep learning ones, usually the SOTA / most recent architectures tend to win if used properly and are, again, ensembled.

Actually Kaggle is not all that far from the real world experience. The problems and used models are similar, if not the same ones. By tackling some competitions, you will learn a lot of practicalities about the domains that are topics for the competitions and you will get a sense of what could work and couldn’t in a specific situation. You will be shown the best performing technicalities and a lot of ideas to be applied in real world problems just by participating to a single competition!

as for as the progression in Kaggle rankings, the progression system has some requirements: https://www.kaggle.com/progression, thus for instance it takes at least a few competitions with the right medals to achieve the level you want. Said that, in my experience is just a matter of time and efforts you put on it. The more time and efforts, the less it will take as a general rule but the exact time depends also on your starting level of skill, how fast you learn, and also if you have good machines at your disposal for fast experimentation and prototyping. Since you learn by doing, also doing faster (because your machine can train or process the data in a faster way) will accelerate your learning.

Luca Massaron Thank you.

Hi Michal, I am super glad to hear that you enjoyed our book and you found it useful for you! Great!
Coming to your questions:

1. some competitions can be cracked using just the Kaggle notebooks or Google Colab. And sometimes even in more complex competitions, such as the M5 (https://www.kaggle.com/competitions/m5-forecasting-accuracy), that apparently seem demanding huge resources, you can actually find top-10 solutions that do not need more that the resources that Kaggkle freely offers to you. Don’t get stopped by not having a powerful computer at home or not wanting to rent a machine on the cloud!
   
2. as in real world problems, automl can be useful in certain Kaggle competitions, for instance for blend or stacking. The very problem is anyway that, if automl works well in a competition, you cannot expect you being the only one having access to it. When a resource is freely available around, it cannot provide much competitive advantage!

Hi Tânia Ferreira!
Actually I learned so much from Kaggle, it is difficult to write everything down! Generally speaking, I can say that above all, Kaggle helped me to have a more practical approach to problems, to experiment faster and in a more effective way, to figure out solutions or methods that I wouldn’t have considered or known otherwise just by reading books and by self-experience.

As for as the porfolio question, yes definitely, especially if the Kaggle competition is on a problem related to the company you are presenting for being hired and if you can demonstrate having done some original and careful work in that challenges. There is no need having done many challenges or having scored very high in the leaderboard, all you bneed to demonstrate by such a portfolio is your passion for data science, your skills in coding a solution, your attention and efforts on a specific problem.

Luca Massaron thank you very much :)

Hi Akmal Azzam, you can the detailed list of contents, together with a sample of the first chapter, here:
https://www.packtpub.com/product/the-kaggle-book/9781801817479

Thank You for Sharing!

Hi Carlos Orjuela! I can understand how you feel. My start at Kaggle has been a real disaster and it took time before I could climb the rankings. Naturally (I didn’t know that at start) you first need to dive deep before climbing up (because you have to learn how to compete, something we had in mind when we wrote the Kaggle Book)! Of course seeing your rank being not so high can be disappointing and may prevent you to invest more time and efforts on improving it. But wait a moment, that could be a problem if you just focus on rankings. In reality rankings are relative, ephemeral and often they are not so much considered out of Kaggle. Instead, focus on the fact that learning is the key. Try participating for learning, just take part in discussions and notebook building and don’t care too much in rankings for a while. Very soon you will have gathered enough competiting skills and knowledge that you will rank higher in competitions without much effort. Just refocus on learning for now!

Awesome, much appreciated 🙂 Luca Massaron

Hi Bharat and thank you a lot for your words!
As for as your questions, I have to say that:

• Kaggle resembles a part of real-life data science: the problems you tackle are real ones, the models you use in competitions are what you would use also in real settings and all the problem solving skills you get from participating in Kaggle are solid and true in every situation!

• Even if you don’t know much of the domani knowledge, you can always keep up by reading the forums and researching blogs and papers on the topic. That’s should be enough even for very specialistic domains! Ah, and forget to look for similar Kaggle competitions, you find a lot by reading them!

Hi Evren Unal! They provide a notebook environment, there you can use the debugging strategies that are commonly used with notebooks and colab.

Hi insop and thank you for your compliments! Since I work a lot with tabular data, I would suggest the Tabular Playground Series (https://www.kaggle.com/search?q=tabular+playground+series), but I am a bit biased for that! What is your work or research interest at the moment?

Hi A Joseph, thank you for kind words! When I had the chance to use a Kaggle dataset for a problem (or as an example when writing a book’s chapter) I tended to evaluate its quality directly, by putting on the table many considerations in respect of the purpose I need it for. Therefore I often made some EDA from data and tried to build a basic model to check if the dataset really answered to my needs! Another aspect to care about when using Kaggle datasets is the fidelity and if the data is maintained. Fidelity is an issue if the data has an original source we expect it represents (in many cases the dataset is a mirror of another source). Maintenance is instead necessary if the data is representing some on-going fenomena (such as a time series, for instance).

Cool. Thanks a lot 👍🏼

Hi Prashant Choudhary! In the Kaggle book we have an entire chapter dealing with that (blending and stacking), I cannot but refer to that specific chapter for figuring why, when and how to ensemble models!

p.s. we did a lot of research on that, you can find all the state of the art written down 🙂

Awesome!🙂

Hi Muhammad Awon! At a certain point the book was growing so thick that we had to stop! I regret not being able to put into it all the interview I wanted and not having space for chapters showing how to practically reproduce Kaggle solutions to most renown competitions. By replicating top solutions you can learn really a lot and it can accelerate your becoming a great Kaggler!

The Kaggle Book has many facets. Yes, it can be used as a text for applied machine learning, too 🙂 You won’t find any other book around with so extensive chapters on practical topics such as ensembling or doing proper cross validation.

Very insightful. Thank you.

As an Ex-chef myself, we share the same interest 😃

Any field can get advantage from machine learning, actually, you just need to have the imagination to fit it into any existing process or figuring out new processes leveraging it!

It depends on the purpose I have for that competition. If I aim for a top ranking, that could imply spending all my free time for a long time, one month, one month and a half. If instead I just plan to learn from the competition, by trying refactoring some notebooks or implementing some ideas of my own, I could spend as little as 4-8 hours on the competition.

thank you