No. 380 of 500 No. 59 in Hands-On AI & Programming

Approaching (Almost) Any Machine Learning Problem

By Abhishek Thakur

Published
June 30, 2020
Print length
300 pages
Reading level
Intermediate
ISBN-13
978-8269211504
Publisher
Abhishek Thakur
Written for
Practitioners with basic ML theory who want a competition-tested workflow for real projects.

Synopsis

Abhishek Thakur, who holds Kaggle's Grandmaster rank in all four of its categories, wrote this self-published book to show how an experienced competitor sets up and works through a machine learning problem. It does not teach algorithms from first principles. It assumes readers already know the theory and concentrates on the decisions that drive results: setting up a working environment, choosing a cross-validation scheme that fits the data, selecting evaluation metrics, structuring project code, and handling categorical variables, feature engineering and feature selection. A chapter on hyperparameter optimization moves from grid and random search to model-based tuning libraries.

Later chapters apply the same workflow to image classification and segmentation, text classification and regression, and ensembling and stacking, then close with reproducible code and model serving. The book is dense with Python code, has no subheadings, and expects readers to type the code themselves; the author does not distribute it separately. Thakur released the full PDF free on GitHub, where the repository has drawn more than 8,000 stars, and sells print and Kindle editions. Readers come away with a reusable template for tackling tabular, image and text problems and a sharper sense of where validation mistakes undermine results.

Scorecard

Editorial score

How it scored

Editorial score: 70.1 / 1005 weighted criteria

  • Influence30%

    21.0/30
  • Authority25%

    18.8/25
  • Practical value20%

    16.0/20
  • Currency15%

    8.3/15
  • Readability10%

    6.0/10
  • Editorial score

    70.1/100

How the rank was computed

Rank No. 380 of 500 comes from a composite of 50.6: observable signals 19.1 of 55 points plus editorial review 31.5 of 45 points.

All 11 inputs
InputPoints
Observable signals19.1/55
Reach (Amazon ratings volume)10.7/16.5
Consensus across best-of lists2.8/13.8
Awards and bestseller evidence0.0/8.3
Reader rating, sample-adjusted5.5/8.3
Amazon sales rank0.1/5.5
Recency0.0/2.8
Editorial review31.5/45
Author authority8.4/11.3
Influence9.5/13.5
Practical value7.2/9.0
Readability2.7/4.5
Currency3.7/6.8

Each criterion is scored from 1 to 10 and weighted, and the points add up to the editorial score. The rank combines observable signals (55 percent) with this score (45 percent). Read the methodology.

Verdict

A code-first field manual from a top Kaggle competitor, strongest on practical workflow and weakest on explaining the algorithms themselves.

Strengths

  • Practical workflow distilled from Kaggle competition experience
  • Strong treatment of validation and metrics
  • Full PDF freely available on GitHub

Keep in mind

It explains little theory, lacks subheadings, and requires typing code from the page, so beginners and readers wanting algorithm explanations may struggle.

Key takeaways

  1. Thakur treats a cross-validation scheme matched to the data, such as stratified or grouped folds, as the foundation of every reliable experiment.
  2. Evaluation metrics should match the problem; accuracy can mislead on imbalanced data, where precision, recall, F1 or AUC are more informative.
  3. Categorical variables can be handled through label encoding, one-hot encoding, target encoding or learned entity embeddings, each suited to different models.
  4. Ensembling and stacking several diverse models usually outperforms a single tuned model, provided the blending is validated out of fold.

Notable facts

About the author

Abhishek Thakur

Abhishek Thakur holds Kaggle's Grandmaster rank in all four of its categories: competitions, datasets, notebooks and discussions. He has worked on natural language processing at Hugging Face and publishes coding tutorials on his YouTube channel. He self-published this book in 2020 and later released the full text free on GitHub.

Questions

Is Approaching (Almost) Any Machine Learning Problem good for beginners?
Not for complete beginners. The author states that the book does not explain the algorithms and assumes readers already understand basic machine learning and deep learning theory. It suits learners who have taken an introductory course and now want to see how an experienced competitor validates models, engineers features, tunes hyperparameters and organizes code on real problems.
Is Approaching (Almost) Any Machine Learning Problem free?
Yes. Abhishek Thakur released the complete book as a free PDF in his GitHub repository, approachingalmost, while continuing to sell paperback, color and Kindle editions. The author warns Indian buyers that some paperback copies sold online are unauthorized printouts and points them to the official publisher, Pothi. The code is not distributed separately because the author wants readers to type it themselves.
Is Approaching (Almost) Any Machine Learning Problem still relevant in 2026?
Its core lessons on cross-validation, metrics, categorical encoding, feature selection and ensembling remain sound for tabular and classic ML work. The book has had no new edition since 2020, so its deep learning code predates current library versions and it does not cover large language models or modern fine-tuning. Use it for workflow discipline and supplement it with recent resources on foundation models.

Sources

As an Amazon Associate, Best AI Books earns from qualifying purchases. Best AI Books is sponsored by Iternal.ai, whose founder and CEO, John Byron Hanby IV, wrote The AI Strategy Blueprint, our #1 ranked book. How we rank books.