Leanpub Header

Skip to main content

Topics

Machine Learning

From regression, decision trees and the bias and variance of a model to neural networks, Transformers, interpretation and the running of models in production, machine learning as told through the free samples of the Leanpub books about it.

Created from the free samples of 23 Leanpub books and the descriptions and tables of contents of 18 more books and courses, with every source cited.

Machine learning is the part of computer science that builds programs whose behaviour is derived from examples rather than written out by hand. The best-selling book on this shelf defines it twice: as the study of algorithms that depend on a collection of examples of some phenomenon, and as a process, gathering a dataset and then building a statistical model from it to solve a practical problem. Its preface adds a caution about the name. A learning machine finds a mathematical formula that gives the desired outputs for its training inputs and for others drawn from a similar distribution, and a slight distortion of the input can make the output wrong, which is not how an animal learns.[1]

The field now runs from a line fitted to a table of figures to language models with billions of parameters, and its books disagree about where the weight should fall. One guide declines to use the term artificial intelligence at all, finding it broad and vague, and notes that many people who say it mean deep learning.[2] Another warns against reaching for a deep network where an older model would serve, since the classical models still dominate industrial controllers and often perform as well with less trouble.[9] Around the algorithms a second literature has grown, about deciding whether a problem suits machine learning at all and about carrying a model from an experiment into a system that has to be maintained.[7][5]

This article covers where to begin, the history, what learning means and how a model is judged, the classical methods, neural networks and deep learning, language models, the engineering of machine learning systems, interpretation and uncertainty, and machine learning in particular industries and languages. Its sources are the free samples of the Leanpub books listed at the end, with the descriptions and tables of contents of the books and courses whose samples were not available, and every claim that rests on a source is cited to it.

Where to start

A reader with no background has three short books to choose from. The Hundred-Page Machine Learning Book (2019) assumes no advanced mathematics or programming, gives its notation a chapter of its own and is distributed on what its author calls a read first, buy later principle.[1] Zefs Guide to Deep Learning explains the ideas of machine learning and deep learning in prose, without code,[2] and Zefs Guide to Deep Learning Flashcards is a set of review cards that accompanies it.[26] Machine Learning Algorithms For Laymen is a primer of some fifty pages meant to be read before anything longer.[25]

A programmer who learns by building can start from code. Hands-On Machine Learning from Scratch (2019) writes the classic algorithms by hand in Python, each for a small problem;[4] Machine Learning Pipeline (2019) follows a project from the cleaning of data to a first model;[27] Simplifying Machine Learning with PyCaret reaches working models through a low-code library and assumes Python but no knowledge of the subject;[11] and Deep Learning with PyTorch Step-by-Step teaches deep learning in small steps to beginners.[17] Those who work in other languages have Machine Learning for C# Developers Made Easy for C#,[12] Hacker's Guide to Neural Networks in JavaScript (2020) for JavaScript[31] and Behavior Analysis with Machine Learning and R for R.[30]

Readers who already train models are served by narrower books. The Hundred-Page Language Models Book asks for Python, and finds college-level mathematics helpful, on the way to a Transformer written from scratch;[3] A Hands-On Guide to Fine-Tuning Large Language Models with PyTorch and Hugging Face expects deep learning, PyTorch and the vocabulary of attention and embeddings;[18] Interpreting Machine Learning Models With SHAP is for those who know machine learning and want to explain a model's predictions;[15] Machine Learning Engineering (2020) is for engineers and teams taking a project from a business problem to a monitored system;[7] and Statistical foundations of machine learning: the book is written at the level of a master's or doctoral course.[14] Process engineers have a series of their own, beginning with Machine Learning in Python for Process Systems Engineering, which assumes experience of neither Python nor machine learning.[8]

History

The idea of a machine that improves with experience is as old as computing. A program that learned to play draughts and the perceptron, a single artificial neuron trained on examples, both date from the 1950s. Decision trees, support vector machines and ensembles of trees became the working tools of the 1990s and 2000s, while neural networks, revived in the 1980s by the backpropagation algorithm, waited for data and hardware. From 2012 deep networks trained on graphics processors took over the recognition of images and speech, and the Transformer architecture of 2017 led to the large language models of the 2020s.

The books add their own dates and judgements. The term machine learning was coined in 1959 by a pioneer of computer gaming at IBM, and The Hundred-Page Machine Learning Book holds that the company used it as marketing, to attract clients and employees, much as it later promoted the phrase cognitive computing.[1] Zefs Guide to Deep Learning describes two long-standing approaches to artificial intelligence, one built on hand-coded rules and one on learning, of which deep learning is the present form, and attributes the recent spread of machine learning to more data, cheaper storage and computing, and the open-source software that followed from them.[2] The foreword to The Hundred-Page Language Models Book, by a researcher who came to language modelling two decades ago by way of data compression, recalls that counting short sequences of words was a very simple idea and very hard to beat.[3] Statistical foundations of machine learning: the book first appeared in 2004 as a university course companion, written because the textbooks of the time, in its author's view, underplayed statistics and passed over the assumptions made when a model is inferred from data.[14] Interpreting Machine Learning Models With SHAP gives a chapter to the history of its method, from one mathematician's pursuit of a fair division, through its early use in machine learning, to the many variants since.[15]

What learning means

The standard account has four steps. A dataset is collected, the structure of a model is chosen, a loss function is defined that measures how far the model's outputs lie from the known answers, and the loss is minimised by adjusting the model's parameters. The Hundred-Page Language Models Book sets the steps out with a line fitted to the areas and prices of three houses, solved by hand, and observes that the hand method stops being feasible long before a model has billions of parameters.[3] Zefs Guide to Deep Learning uses the price of a house too, and describes regression as finding the slope and intercept of a best-fit line by trying values and comparing the result with the known data, which it calls the core of the subject: learning the best parameter values of a model from existing data.[2]

Kinds of learning

In supervised learning each example carries a label, and the model learns to produce the label from the example's features. The Hundred-Page Machine Learning Book describes an example as a feature vector, each position of which always holds the same kind of information, and works through the detection of spam: ten thousand labelled messages, each turned into a vector of twenty thousand ones and zeros recording which words of a dictionary it contains, and an algorithm that finds the boundary separating the two classes.[1] Predicting a number is regression and predicting a category is classification, and Zefs Guide to Deep Learning likens a learned classifier to a doctor's diagnostic flow chart whose thresholds have been found from data instead of being written down by the doctor.[2]

Unsupervised learning has no labels and looks for structure: clusters of similar items, anomalies, a smaller set of features that preserves what matters.[1][12] The groups a clustering finds can be named only after their members have been examined, which is part of what makes the method harder.[2] Between the two lie semi-supervised learning, in which many unlabelled examples improve a model trained on a few labelled ones because a larger sample better reflects the distribution the data came from,[1] and self-supervised learning, in which the answer is already in the data, as when a model is trained to predict the next word of a sentence.[2]

Reinforcement learning is the fourth kind. A machine acts in an environment, is rewarded or not, and learns a policy, a rule for choosing the action in each state that maximises the expected reward; it suits problems in which decisions come in sequence, such as games, robotics and logistics.[1] Both introductory books leave it out of their scope,[1][2] and the C# guide notes that its library does not support it.[12] Machine Learning in Python for Process Systems Engineering is the exception on the shelf, with a chapter that ends in an agent trained to control the level of liquid in a tank.[8]

Judging a model

A model is judged on data it has not seen. Zefs Guide to Deep Learning separates two kinds of error: bias, the systematic error that remains when the same kind of model is trained on different samples of the data, and variance, the spread of those models' predictions for a single input. A model that overfits has learned the particular patterns of its sample and so varies widely, and in practice the best choice trades a little of one error for a little of the other. The same chapter recommends setting a baseline before any training, by measuring the existing solution or building the simplest possible one, because machine learning often needs a great deal of data and time before it performs well and a poor idea is otherwise discovered late.[2] The Hundred-Page Language Models Book makes a related point about the loss itself: whether a value is high or low depends on the business context and on what it is compared with.[3]

The ways of getting this wrong have a book to themselves. MOST COMMON MISTAKES IN MACHINE LEARNING AND HOW TO AVOID THEM: With Examples in Python collects seventeen errors its author has met while grading assignments, supervising students and reading published code, among them reporting performance on the training data, tuning on the test set, leaving out a baseline and letting information leak from the answers into the features.[28] The contents of Machine Learning Engineering name three causes of such leakage: a target that is a function of a feature, a feature that hides the target, and a feature that comes from the future.[7]

Classical methods and tabular data

The algorithms that preceded deep learning remain the first resort for data that fits in a table. The chapter of The Hundred-Page Machine Learning Book on fundamental algorithms covers linear regression, logistic regression, decision trees and support vector machines, each as a problem statement followed by a solution.[1] Hands-On Machine Learning from Scratch implements such models from nothing. Its sample builds k-means clustering, which picks centres at random, assigns every point to the nearest centre, recomputes each centre from its points and repeats until the centres stop moving, and uses it to extract the colour palette from screenshots of mobile applications; the next chapter classifies film reviews as favourable or not with naive Bayes.[4]

Much of the work lies before and around the algorithm. Hacker's Guide to Machine Learning with Python (2020) opens with a model that predicts the nightly price of a rented room from some forty-nine thousand listings, and insists on exploring the data first; finding the prices heavily skewed, it takes their logarithm before training.[10] Machine Learning Pipeline gives data types and levels of measurement a chapter of their own, though its contents stop at logistic regression.[27] Feature Selection in Machine Learning is devoted to choosing which columns a model should see, with more than twenty methods of the filter, wrapper and embedded kinds.[29] The Orange Book of Machine Learning - Green edition confines itself to supervised learning on tabular data, from descriptive statistics and exploratory analysis through regression, classification and ensembles to the recent foundation models for tables.[16] Simplifying Machine Learning with PyCaret shows the opposite economy, a library in which regression, classification, clustering, anomaly detection and forecasting each take a few calls, with a final chapter on publishing a model as a small web application.[11]

Statistical foundations of machine learning: the book is the theoretical counterpart. Its author came to see machine learning as bound up with the scientific process of turning observations into knowledge, and the book runs from probability and estimation through linear and nonlinear models to causal inference, with code in R.[14] Two adjacent books sit at the edge of the subject. Genetic Algorithms with Python (2019) teaches evolutionary search, which needs only a measure of fitness that says which of two guesses is better, and explains the idea with a number-guessing game that becomes solvable once the feedback is higher or lower instead of right or wrong.[19] A Mathematical Theory of the Unknown is a long mathematical treatise on measuring how much is not known about a subject, built on the theory of computability.[41]

Neural networks and deep learning

A neural network is one kind of machine learning model: a collection of mathematical operations that turns raw input into a prediction, with parameters learned from training data. Deep learning, in the words of Zefs Guide to Deep Learning, is a name for the modern use of deep neural networks, and its later chapters proceed to the convolutional networks used for images, to sequence models and to Transformers.[2] Deep Learning with PyTorch Step-by-Step covers the same ground at length and in code, in volumes on fundamentals, computer vision and sequences, and begins by visualising gradient descent, the procedure by which a network's parameters are adjusted.[17] Understanding Deep Learning ties its account of how the constructs of deep learning behave to one application, the prediction of rare events.[33]

Several books teach by project. Get SH*T Done with PyTorch (2020) works through problems in vision, language and time series and ends by serving a model through a web interface;[32] Hacker's Guide to Machine Learning with Python goes on from its tabular example to recurrent networks, object detection and a pretrained language model;[10] and Hacker's Guide to Neural Networks in JavaScript builds networks from scratch and then trains and runs them in the browser.[31] Machine Learning in Python for Visual and Acoustic Data-based Process Monitoring applies convolutional networks to two industrial tasks, finding defects in images of products and detecting abnormal sounds from equipment.[34] The older tradition of image analysis is represented by an adjacent book, Fundamentals of Computer Vision (2020), a guide written beside a university course that covers filtering, edges, cameras, tracking and recognition and states plainly that it leaves deep learning out.[20]

Language models

A language model, The Hundred-Page Language Models Book says, is at its core a mathematical function, one whose abilities are best appreciated by implementing it. The book deliberately includes the approaches that modern accounts skip, models that count sequences of words and recurrent networks, before writing a Transformer in Python and turning to fine-tuning and the use of large models; it says equally plainly what it omits, namely training at scale, deployment and enterprise applications.[3] The same material is published as a course with quizzes, exercises and recorded interviews.[24]

Three shorter books explain the same machinery with less mathematics. The inner workings of Large Language Models follows a piece of text from tokens and embeddings through attention to the choice of the next token;[35] Super Study Guide: Transformers & Large Language Models is an illustrated study guide in five parts, written partly for interview preparation;[36] and LM from First Principles grows a model that predicts one character from the last into something larger, adding one capability a chapter.[37] At the other end, Build an LLM Inference Engine in C++ has C++ programmers write an engine that loads an open model and generates text on an ordinary processor,[38] and A Hands-On Guide to Fine-Tuning Large Language Models with PyTorch and Hugging Face concentrates on the configuration settings that decide whether a large model can be adapted on consumer hardware.[18]

Two books apply the models to particular material. Generative AI for Science grew out of a graduate course and pairs each chapter with notebooks on chemistry, biology, physics or climate data; its first chapter places generative models after the microscope and genome sequencing as instruments that change what science can do, citing the prediction of protein structures, and its front matter sets out in detail how AI assistants were used in writing it.[13] Introduction to Japanese Natural Language Processing, an adjacent book, is about processing Japanese text, and its sample shows a tokeniser that returns each word with its part of speech, dictionary form and reading.[23]

From notebook to production

Most models never leave the notebook in which they were trained, and a group of books is about the ones that must. Machine Learning Engineering begins before the project does. Its contents list when machine learning is worth using, for a problem too complex to code, one that changes constantly, one of perception, one that has not been studied, one with a simple objective, and only where it is cost-effective, and then the questions to ask of the data: whether it is accessible, sizeable, usable, understandable and reliable.[7]

Clean Machine Learning Code (2020) describes the moment of transition. A team that has been exploring is abruptly asked for maintainability and reliability, and a feature that took three hours to build as a prototype is estimated at four weeks in production. The author doubts the promise that managed services will make such code unnecessary, argues that machine learning pipelines are software pipelines with software's faults, and names those faults in the vocabulary of software design: code that must be changed in six places at once, that breaks somewhere else when it is fixed, that cannot be reused, and whose names explain nothing.[5] Deep Learning in Production makes the same case constructively by following one image-segmentation model from a notebook to a served application. Its sample chapter lays the project out in packages for configuration, data loading, the model and its execution, turns the model into a class behind an abstract interface with methods for loading data, building, training and evaluating, and advises writing Python as one would write Java.[6]

The surrounding skills have books too. Advanced Python for Data Science (2021), grown from a university course, holds that a data scientist's effect is limited less by ideas than by the ability to enact them: reaching the data, storing it, processing it at scale, running and testing the same script every day, and packaging the result for others.[21] Cloud Computing for Data Analysis (2021) introduces the cloud services, containers and managed training platforms such work relies on,[40] and Hiring Data Scientists and Machine Learning Engineers (2021) is a guide to hiring, written because the field's sudden popularity has left it unclear which skills a role needs while courses and boot camps supply applicants in great numbers.[22]

Interpretation and uncertainty

A model that predicts well may still have to explain itself. Interpreting Machine Learning Models With SHAP is a whole book on one method of explanation, which divides a prediction among the features that produced it. Its contents give the reasons for interpreting a model, to debug it and to build trust in it, set out the limits of models that are interpretable by design, and introduce the underlying theory with the question of how passengers should split a taxi fare, before turning to estimation, to particular kinds of model and to the method's own limitations.[15]

Uncertainty is a separate question: not why the model answered as it did, but how far the answer can be relied on. The Orange Book of Machine Learning - Green edition opens by distinguishing the uncertainty that comes from chance from the uncertainty that comes from ignorance, and confidence intervals from prediction intervals.[16] Applied Conformal Prediction:Practical Uncertainty Quantification for Real-World ML, still being released, treats one framework for attaching statistically valid intervals to the predictions of any model, across classification, regression and time series.[39]

Industries and languages

The largest group of applied books on the shelf comes from the process industries. The preface of Machine Learning in Python for Process Systems Engineering observes that away from the visible products, machine learning has been advancing quietly in chemical, pharmaceutical and steel plants, in predictive maintenance, quality assurance, process monitoring and control, and that its authors could find no data science book written for a process engineer; theirs teaches each method through a plant dataset and ends by deploying a monitoring application.[8] Machine Learning in Python for Dynamic Process Systems continues the series with models of processes that change over time, uses Python where the literature of its field has used MATLAB, and gives the classical models as much room as neural networks.[9] Behavior Analysis with Machine Learning and R draws its examples from human behaviour instead, recognising activities from the sensors of a phone and gestures from movement, and gives attention to validating a model across different users.[30]

Python is the language of nearly every book here, and the exceptions are written for programmers who would rather not change. Machine Learning for C# Developers Made Easy keeps the whole workflow in C# and compares its reader to a driver, who need not understand the engine but must know where the controls are.[12] Hacker's Guide to Neural Networks in JavaScript does the same for JavaScript,[31] and the statistical texts use R.[14][30]

Questions about Machine Learning

Which machine learning book should I start with?

A reader with no background can start with The Hundred-Page Machine Learning Book (2019), which assumes no advanced mathematics or programming and gives its notation a chapter of its own, or with Zefs Guide to Deep Learning, which explains the ideas of machine learning and deep learning in prose, without code.[1][2] A programmer who learns by building can start from Hands-On Machine Learning from Scratch (2019), which writes the classic algorithms by hand in Python, each for a small problem.[4] For deep learning, Deep Learning with PyTorch Step-by-Step teaches in small steps to beginners.[17]

What is machine learning?

The Hundred-Page Machine Learning Book defines machine learning twice: as the study of algorithms that depend on a collection of examples of some phenomenon, and as a process, gathering a dataset and then building a statistical model from it to solve a practical problem.[1] Its preface adds a caution about the name: a learning machine finds a mathematical formula that gives the desired outputs for its training inputs and for others drawn from a similar distribution, and a slight distortion of the input can make the output wrong, which is not how an animal learns.[1]

What is the difference between supervised and unsupervised learning?

In supervised learning each example carries a label, and the model learns to produce the label from the example's features. Unsupervised learning has no labels and looks for structure: clusters of similar items, anomalies, a smaller set of features that preserves what matters.[1][12] The groups a clustering finds can be named only after their members have been examined, which is part of what makes the method harder.[2]

What are bias and variance in machine learning?

Zefs Guide to Deep Learning separates two kinds of error: bias, the systematic error that remains when the same kind of model is trained on different samples of the data, and variance, the spread of those models' predictions for a single input.[2] A model that overfits has learned the particular patterns of its sample and so varies widely, and in practice the best choice trades a little of one error for a little of the other.[2]

What is data leakage in machine learning?

Leakage is information passing from the answers into the features, one of the seventeen errors MOST COMMON MISTAKES IN MACHINE LEARNING AND HOW TO AVOID THEM: With Examples in Python collects, beside reporting performance on the training data, tuning on the test set and leaving out a baseline.[28] The contents of Machine Learning Engineering (2020) name three causes of it: a target that is a function of a feature, a feature that hides the target, and a feature that comes from the future.[7]

When is machine learning worth using?

The contents of Machine Learning Engineering (2020) list the cases: a problem too complex to code, one that changes constantly, one of perception, one that has not been studied, one with a simple objective, and only where it is cost-effective.[7] Zefs Guide to Deep Learning recommends setting a baseline before any training, by measuring the existing solution or building the simplest possible one, because machine learning often needs a great deal of data and time before it performs well.[2] Machine Learning in Python for Dynamic Process Systems warns against reaching for a deep network where an older model would serve.[9]

Sources

Every source is a Leanpub book or course. The note after each says which part of it was read; where no sample was available, only the public description and table of contents were used.

  1. 1.
    The Hundred-Page Machine Learning Book, Andriy Burkov Free sample: the preface, the contents of the first three chapters, and the opening of chapter one on the kinds of learning and how supervised learning works. 2019. Free sample: PDF
  2. 2.
    Zefs Guide to Deep Learning, Roy Keyes Free sample: the chapter on machine learning basics, with tasks, kinds of learning, performance, baselines, bias and variance, and the opening of the chapter on neural networks. 2024. Free sample: PDF · EPUB
  3. 3.
    The Hundred-Page Language Models Book, Andriy Burkov: hands-on with PyTorch Free sample: the contents, the foreword, the preface and the opening of chapter one, on the four steps of machine learning. 2025. Free sample: PDF
  4. 4.
    Hands-On Machine Learning from Scratch, Venelin Valkov: Develop a Deeper Understanding of Machine Learning Models by Implementing Them from Scratch in Python Free sample: the chapter that builds k-means clustering by hand and the opening of the chapter on naive Bayes. 2019. Free sample: PDF · EPUB
  5. 5.
    Clean Machine Learning Code, Moussa Taifi Free sample: chapter one, on why the quality of machine learning code matters and the design problems of a neglected pipeline. 2020. Free sample: PDF · EPUB
  6. 6.
    Deep Learning in Production, Sergios Karagiannakos Free sample: the chapter on structuring deep learning code, with project layout, classes and abstract interfaces. 2022. Free sample: PDF
  7. 7.
    Machine Learning Engineering, Andriy Burkov Free sample: front matter and the contents of the first chapters, on when to use machine learning, the project life cycle and data. 2020. Free sample: PDF
  8. 8.
    Machine Learning in Python for Process Systems Engineering, Ankur Kumar, Jesus Flores Cerrillo: Achieve Operational Excellence Using Process Data Free sample: the full contents, the preface and the notes on readers and organisation. 2024. Free sample: PDF
  9. 9.
    Machine Learning in Python for Dynamic Process Systems, Ankur Kumar, Jesus Flores Cerrillo: A practitioner’s guide for building process modeling, predictive, and monitoring solutions using dynamic data Free sample: the preface, the intended readers, the organisation of the book and its contents. 2024. Free sample: PDF
  10. 10.
    Hacker's Guide to Machine Learning with Python, Venelin Valkov: Hands-on guide to solving real-world Machine Learning problems with Scikit-Learn, TensorFlow 2, and Keras Free sample: the first project, predicting rental prices, from exploring the data to training and saving a model. 2020. Free sample: PDF · EPUB
  11. 11.
    Simplifying Machine Learning with PyCaret, Giannis Tolios: A Low-code Approach for Beginners and Experts! Free sample: the preface, the intended reader and setup, and the start of the regression chapter. 2023. Free sample: PDF · EPUB
  12. 12.
    Machine Learning for C# Developers Made Easy, Fiodar Sazanavets: Build smart applications with ML.NET Free sample: chapter two, on the kinds of machine learning; chapter one appears as headings only. 2026. Free sample: PDF · EPUB
  13. 13.
    Generative AI for Science, J. Paul Liu: A Hands-On Guide for Students and Researchers Free sample: the author's notes, prerequisites and statement on the use of AI, and the opening of chapter one. 2026. Free sample: PDF · EPUB
  14. 14.
    Statistical foundations of machine learning: the book, Gianluca Bontempi Free sample: the preface and the contents of the opening chapters. 2026. Free sample: PDF
  15. 15.
    Interpreting Machine Learning Models With SHAP, Christoph Molnar: A Guide With Python Examples And Theory On Shapley Values Free sample: title pages and the full table of contents. 2026. Free sample: PDF
  16. 16.
    The Orange Book of Machine Learning - Green edition, Carl McBride Ellis: The essentials of making predictions using supervised regression and classification for tabular data. Free sample: the title page and contents of the free edition, covering the introduction, statistics and exploratory analysis. 2026. Free sample: PDF
  17. 17.
    Deep Learning with PyTorch Step-by-Step, Daniel Voigt Godoy: A Beginner's Guide Free sample: front matter and the start of the table of contents only. 2024. Free sample: PDF
  18. 18.
    A Hands-On Guide to Fine-Tuning Large Language Models with PyTorch and Hugging Face, Daniel Voigt Godoy Free sample: front matter and the table of contents only. 2025. Free sample: PDF
  19. 19.
    Genetic Algorithms with Python, Clinton Sheppard Free sample: the preface and the introduction to genetic algorithms and fitness. 2019. Free sample: PDF
  20. 20.
    Fundamentals of Computer Vision, George K: A gentle, accessible introduction to foundational concepts in computer vision and computational perception. Free sample: the preface and the table of contents. 2020. Free sample: PDF
  21. 21.
    Advanced Python for Data Science, Scott Gorlin Free sample: the introduction, on what the book is and who it is for. 2021. Free sample: PDF · EPUB
  22. 22.
    Hiring Data Scientists and Machine Learning Engineers, Roy Keyes: A Practical Guide Free sample: the introduction, on the purpose of the guide and why it is needed. 2021. Free sample: PDF · EPUB
  23. 23.
    Introduction to Japanese Natural Language Processing, Masato Hagiwara, Paul O'Leary McCann Free sample: the opening of the chapter on tokenisation. 2026. Free sample: PDF
  24. 24.
    The Hundred-Page Language Models CourseCourse, Andriy Burkov No sample: from the description; the course version of the language models book. 2025.
  25. 25.
    Machine Learning Algorithms For Laymen, Ashmin, Asal Kc No sample: from a brief description. 2025.
  26. 26.
    Zefs Guide to Deep Learning Flashcards, Roy Keyes No sample: from the description. 2022.
  27. 27.
    Machine Learning Pipeline, Hisham El-Amir: Experience Gain No sample: from the description and a table of contents that stops at logistic regression. 2019.
  28. 28.
    MOST COMMON MISTAKES IN MACHINE LEARNING AND HOW TO AVOID THEM: With Examples in Python, Enrique Garcia-Ceja No sample: from the description and table of contents. 2025.
  29. 29.
    Feature Selection in Machine Learning, Soledad Galli, PhD: Over 20 methods to select the most predictive features and build simpler, faster, and more reliable machine learning models. No sample: from the description and table of contents. 2026.
  30. 30.
    Behavior Analysis with Machine Learning and R, Enrique Garcia Ceja: A Sensors and Data Driven Approach No sample: from the description and table of contents. 2023.
  31. 31.
    Hacker's Guide to Neural Networks in JavaScript, Venelin Valkov: Beginners guide to understanding Machine Learning in the browser with TensorFlow.js No sample: from the description and table of contents. 2020.
  32. 32.
    Get SH*T Done with PyTorch, Venelin Valkov: Solve Real-World Machine Learning Problems No sample: from the description and table of contents. 2020.
  33. 33.
    Understanding Deep Learning, Chitta Ranjan: Application in Rare Event Prediction No sample: from a short description. 2022.
  34. 34.
    Machine Learning in Python for Visual and Acoustic Data-based Process Monitoring, Ankur Kumar: A short beginner’s guide to deep learning-based computer vision and abnormal sound detection No sample: from a short description. 2024.
  35. 35.
    The inner workings of Large Language Models, Roger Gullhaug: how neural networks learn language No sample: from the description and table of contents. 2025.
  36. 36.
    Super Study Guide: Transformers & Large Language Models, Afshine Amidi, Shervine Amidi No sample: from the description alone. 2025.
  37. 37.
    LM from First Principles, Amarpreet Singh Bassan: A Notebook-Driven Guide to Building and Deploying Language Models No sample: from the description and table of contents. 2026.
  38. 38.
    Build an LLM Inference Engine in C++, Hatem M.: A Challenge-Driven Guide to Building a CPU-First Inference Engine in C++20 No sample: from the description and table of contents. 2026.
  39. 39.
    Applied Conformal Prediction:Practical Uncertainty Quantification for Real-World ML, Valery Manokhin: Practical Uncertainty Quantification for Real-World ML Learn Conformal Prediction (CP), the state-of-the-art technique for building statistically valid, model-agnostic prediction intervals No sample: from the description of a book still being released. 2026.
  40. 40.
    Cloud Computing for Data Analysis, Noah Gift: The missing semester of Data Science No sample: from the description and table of contents. 2021.
  41. 41.
    A Mathematical Theory of the Unknown, R. A. García Leiva: Journey Beyond the Frontiers of Human Understanding No sample: from the description and table of contents. 2026.

Further reading

Other books in the library on neighbouring ground, not cited above.

  • Running Local LLMs on Your Own Hardware, Yohan J. Rodríguez: A Practical Guide to Private, Offline, and Self-Hosted Large Language Models Downloading, running and serving open-weight language models on one's own hardware, with chapters on quantisation and fine-tuning. 2026. Free sample: PDF
  • Generative AI with local LLM, Shamim Bhuiyan, Timur Isachenko: A comprehensive roadmap for building AI-Driven applications with local LLMs A project-based path through generative AI applications on locally run models, including a chapter on fine-tuning. 2025. Free sample: PDF · EPUB
  • Building LLM and AI Agent-Based Applications for the Process Industry, Ankur Kumar, Akhilesh Jain: A gentle introduction to building useful agentic AI industrial solutions From the same series as the process industry books above: language model and agent applications for plant operations. 2026. Free sample: PDF
  • LLM Prompt Engineering For Developers, Aymen El Amri: The Art and Science of Unlocking LLMs' True Potential Writing and testing prompts for language models, opening with a history of language processing from n-grams to Transformers. 2024.
  • The Agentic AI book, Dr. Ryan Rad: From Language Models to Multi-Agent Systems An engineering guide that runs from language models and fine-tuning to agents and their deployment. 2026.
  • JupyterLab 101, Michael Driscoll: Learn the Latest Iteration of Jupyter Notebook A guide to the notebook environment in which most of the code in these books is run. 2025.

This is an article created from the free samples of 23 Leanpub books and the descriptions and tables of contents of 18 more books and courses. We're building these articles to provide you, the Leanpub reader, with a good jumping-off point to discover more Leanpub books that are interesting and relevant to you. We never use material which is paid-only when building these articles: we only use material which is already free on Leanpub. However, Leanpub authors can opt out from inclusion in articles from their book settings page.

Updated November 2026.

In the Leanpub store: 345 books · 3 courses in the Machine Learning category · also filed under Deep Learning, Neural Networks, Computer Programming/Artificial Intelligence/Python, Machine Learning Systems, Decision Model

All topics