Approaching (Almost) Any Machine Learning Problem Paperback – June 30, 2020

Kaggle has a strange reputation within the data science community. On one hand it's a great source of innovation in a range of sub-fields and when solving a similar problem to an existing Kaggle competition seeing how it was approached by high ranking teams is very valuable. On the other it is a distorted version of what data science actually is in the real world. Usually the (clean) data is provided to you in Kaggle whereas sourcing, collecting and cleaning data is normally a big chunk of a working data scientists life. Finally the approaches in Kaggle competitions are often all about squeezing that tiny improvement out of large numbers of ensembled models. In the real world concessions towards speed, simplicity and interpretability have to be made.The author Abhishek Thakur was the first to achieve GM level across all 4 categories on Kaggle (competitions, kernels, datasets and discussion) . Even a single GM level is an exceptionally difficult task requiring immense amounts of time and skill. My worry going into this book was who it was aimed at and what its purpose is; is it just about doing well on Kaggle or will people who work in industry learn something valuable? Is it aimed at advanced modellers who are looking to become truly elite or would someone with a more general background gain useful knowledge?I am pleased to state that this is a book which is very valuable for the working data scientist and the keen Kaggler. The real value is how it allows us to see how a highly skilled predictive modeller approaches new problems. The book is made up of 13 chapters;ch 1 - Setting up your working environmentch 2 - Supervised vs Unsupervised Learningch 3 - Cross-validationch 4 - Evaluation Metricsch 5 - Arranging Machine Learning Projectsch 6 - Approaching categorical variablesch 7 - Feature engineeringch 8 - Feature selectionch 9 - Hyperparameter optimisationch 10 - Approaching image classification & segmentationch 11 - Approaching text classification/regressionch 12 - Approaching ensembling and stackingch 13 - Approaching reproducible code & model servingWhile Kaggle is great at discussing a highly placed final entry, the value of this book is a walk through of the steps taken towards a solution; ch 1 on Setting up your working environment, ch 5 Arranging a machine learning project and ch 13 on Reproducible code and model serving I found particularly valuable learning some neat tricks on laying out a project. These are all valuable topics which can get lost when we ask "How did the solution work" as in reality the final answer involved lots of iteration lost in just seeing the final product. I really liked the way the author's projects were laid out using a config file and model_dispatcher file allowing for quick modification of which algorithm to use. I had not come across this before but it's a great idea which speeds up the iteration of the modelling process. The other elements that I think makes this book such a great learning solution for people beginning their data science journey is that it shows mistakes which the author then discusses in depth. Finally we see many examples of simpler models beating more complex ones a lesson that is hard to accept when you are starting out and keen to apply XGBoost or Neural Networks to all the things.This book is not just for beginners however - even as someone who has worked as data scientist in industry for a number of years I learnt a great deal from the chapters on dealing with categorical variables, feature engineering and feature selection. As the author notes there are other sources for these solutions but they are spread out across numerous blog posts and forums - having them in a book makes things much easier. I work in customer analytics so when building predictive models a lot of time in spent on feature engineering and feature selection - I learnt a couple of tricks which will be valuable for new projects at work. The book even includes a section on using embeddings on tabular data - a neat approach not widely used in my experience.Finally the book amazingly includes chapters on computer vision problems using PyTorch for classification and image segmentation and nlp using a range of approaches of increasing complexity from bag of words through word2vec to and LSTM and finally a BERT model. The author rightly skips over the complexities of how a CNN or LSTM and Transformer work, but gives enough of a description to get a sense of what is going on. Again the author emphasises the valuable lesson of starting with simpler models and approaches and only then increasing the complexity with constant comparison to a baseline. The author hints (perhaps jokingly) he is considering work on similar books on Computer Vision and NLP - I hope the success of this book encourages him to seriously consider doing this.It is an amazing achievement that the author has created a book which allows the reader to build strong models in a such broad range of domains. The book is well written with the code in particular being excellent. There were one or two spots where the written phrasing was a little hard to follow but these were rare and overall I enjoyed the writing style. The book is eminently practical so the reader will need to find other sources for the theoretical workings of the algorithms used as they gain more experience. Given the breadth the book achieves this is perfectly acceptable. Finally a small technical issue I had with the Kindle version was the lack of a table of contents accessible via the Kindle menu. Not a big thing but does make navigating the book a little trickier than it needs to be.Overall this is an excellent book full of hard won wisdom from a very talented data scientist and educator. I would happily have paid 4 or 5 times its current price and still been very happy with my purchase. I will be highly recommending this book to friends and colleagues who work (or hope to work) in the field.

Abhishek Thakur's "Approaching (Almost) Any Machine Learning Problem" is a game-changer for anyone navigating the complex world of machine learning. As someone relatively new to the field, I found this book to be an invaluable resource that demystifies the process of approaching and solving machine learning problems.Thakur's writing style is clear, concise, and approachable, making complex concepts accessible even for those with a limited background in the subject. The book takes a hands-on approach, providing practical insights and real-world examples that bridge the gap between theory and application. I particularly appreciated the step-by-step guidance on problem formulation, data preprocessing, feature engineering, model selection, and evaluation.One standout feature of the book is its emphasis on a structured and systematic methodology for problem-solving. Thakur's framework not only helps in understanding the nuances of different machine learning tasks but also instills confidence in tackling challenges head-on. The inclusion of case studies further enhances the learning experience, allowing readers to see the methodology in action across various domains.The book strikes the perfect balance between depth and accessibility. Whether you're a beginner looking to build a solid foundation or an experienced practitioner seeking to refine your approach, "Approaching (Almost) Any Machine Learning Problem" caters to a wide audience. The inclusion of practical tips and common pitfalls adds a layer of real-world wisdom that is often missing in other technical books.In conclusion, Abhishek Thakur has created a masterpiece that should be on the bookshelf of every aspiring data scientist or machine learning enthusiast. This book is not just a guide; it's a mentor that equips you with the tools and mindset needed to tackle (almost) any machine learning problem successfully. Highly recommended!

This is a solid book. It’s a bit opinionated, but he does give good reasons. I’m still not wholly convinced about separating the feature engineering from modeling. This is why sklearn pipelines were created, so there’s no chance of bias especially when you do k-fold cross validation. This book is built upon separation of the folds initially. I also was surprised he didn’t use SkLearn Transformers...it’s good overall and I really need to go this over again to really be convinced (or not). If the author reads this, I’d like to hear your comments

This book has a lot of excellent code snippets for approaching various ML problems from tabular supervised learning, NLP, and computer vision. The code is very clean and high quality. Chapter 5 on organizing ML projects is great; this is not a topic often covered by other authors but very important in practice. This book is a very good bridge for someone who has learned some theory and/or taken a MOOC already and is interested in proper application.

Abhishek Thakur's book, Approaching (Almost) Any Machine Learning Problem, is a great resource for anyone who wants to learn more about machine learning. The book is divided into four sections: data preprocessing, model selection, training, and tuning. Each section contains several chapters that cover different topics in detail. The book also includes a section on resources, which includes links to websites and software that can be used for machine learning.Overall, I found the book to be well-written and easy to follow. The book is packed with information, but it is presented in a way that is easy to understand. I would recommend this book to anyone who is interested in learning more about machine learning.

It is an awesome ML book. If you read few ML books and you are not sure ML workflow, what ML model to choose, how to do it? Then this book is the one you need. It can help you put everything you learned ML together. So it is not a beginner book. Thanks.

Many useful tips, i never figured out by my self, worth every cent

This is one of the best hands-on ML books that I have read so far. The way that the author explains make it easy to understand. Also, I liked a lot the coding style and the ML arrangement shown in the book.

Just finished reading the book "Approaching (Almost) Any Machine Learning Project" recently published by Abhishek Thakur .I must admit that i was positively surprised by the atypical content/approach Abhishek implemented in his book.Some of the things that i found really nice:- he motivates and stresses best practices for building and evaluating machine learning models for the most common problems,- he shares different tips & tricks you'll generally spend a lot of time learning while implementing concrete projects / kaggle challenges,- he advocates for a greedy approach towards solving different machine learning problems (always set a clear objective/target and start simple); Abhishek demonstrated that all along the book while discussing different problems.You'll not find in depth explanation of ML algorithms in this book but instead, very good advices on how to properly leverage and apply them in order to solve concrete problems.Thanks a lot Abhishek for the amazing work!

This book is exactly what it claims . It is approach to almost any kind of machine learning problem . The book has a very practical approach towards problem solving and has tones of code . The book is intended to people who knows a bit of background and struggles to start problem solving . The book was and is very helpful for me in day to day life and hence the 5 stars .I want to be clear in the review so that people dont get disappointed after buying the book or refrain from buying the book seeing few negative reviews .The book is immensely helpful for senior professionals like me who are from other branches of software engineering /domain knowledge and still wants to reduce the entry level barrier to get into machine learning and deep learningThe book is immensely helpful for students from programming/non-programming background looking to learn and solve AI/ML problems and are scared to venture into it .The book is immensely helpful for people with a lot of theoretical knowledge but less programming knowledge to apply the same into some cool project .The book is not helpful for people who has a masters in mathematics and looking for reference in writing phd thesis in Machine Learning methods .