What do you think?
Rate this book
Adaptive Computation and Machine Learning
Elements of Causal Inference: Foundations and Learning Algorithms
A concise and self-contained introduction to causal inference, increasingly important in data science and machine learning.
The mathematization of causality is a relatively recent development, and has become increasingly important in data science and machine learning. This book offers a self-contained and concise introduction to causal models and how to learn them from data. After explaining the need for causal models and discussing some of the principles underlying causal inference, the book teaches readers how to use causal models: how to compute intervention distributions, how to infer causal models from observational and interventional data, and how causal ideas could be exploited for classical machine learning problems. All of these topics are discussed first in terms of two variables and then in the more general multivariate case. The bivariate case turns out to be a particularly hard problem for causal learning because there are no conditional independences as used by classical methods for solving multivariate cases. The authors consider analyzing statistical asymmetries between cause and effect to be highly instructive, and they report on their decade of intensive research into this problem.
The book is accessible to readers with a background in machine learning or statistics, and can be used in graduate courses or as a reference for researchers. The text includes code snippets that can be copied and pasted, exercises, and an appendix with a summary of the most important technical concepts.
The mathematization of causality is a relatively recent development, and has become increasingly important in data science and machine learning. This book offers a self-contained and concise introduction to causal models and how to learn them from data. After explaining the need for causal models and discussing some of the principles underlying causal inference, the book teaches readers how to use causal models: how to compute intervention distributions, how to infer causal models from observational and interventional data, and how causal ideas could be exploited for classical machine learning problems. All of these topics are discussed first in terms of two variables and then in the more general multivariate case. The bivariate case turns out to be a particularly hard problem for causal learning because there are no conditional independences as used by classical methods for solving multivariate cases. The authors consider analyzing statistical asymmetries between cause and effect to be highly instructive, and they report on their decade of intensive research into this problem.
The book is accessible to readers with a background in machine learning or statistics, and can be used in graduate courses or as a reference for researchers. The text includes code snippets that can be copied and pasted, exercises, and an appendix with a summary of the most important technical concepts.
288 pages, Hardcover
Published November 29, 2017
46 people are currently reading
257 people want to read
About the author
Jonas Peters
7 books1 followerJonas Peters is Associate Professor of Statistics at the University of Copenhagen.
Ratings & Reviews
Friends & Following
Create a free account to discover what your friends think of this book!
Community Reviews
Displaying 1 - 5 of 5 reviews
February 18, 2018
This book provides a nice introduction into today's causal inference research. For a person like me who is vaguely interested in the topic, but 1) find classical writings like Pearl's to be difficult to understand because they are not written in the language of modern statistics & machine learning, and 2) want to get an overview of today's rapid & diverse research on the topic, this book is a perfect fit. Authors explain key ideas of causal inference in modern terminologies of machine learning, and I found it much more readable than others. They also cover a wide spectrum of ongoing approaches and issues in the field, and make insightful connections between them. Since the book covers so many topics, however, most topics are only sketchily touched, and technical proofs are mostly left out. Moreover, authors concentrate mostly on theoretical issues (ex: identifiability) and applications to real-world problems are only occasionally discussed. This book only serves as a starting point, and you need to follow references to really understand any topic; I expected deeper and gentler dive, at least for key concepts. I also found latter half of the book to be not as carefully written as in the beginning; so many parentheses and hyphens, which are quite distracting.
August 16, 2018
After reading "The Book of Why", I was looking for a more technical introduction to causality. Since by background in machine learning using kernel methods, this book co-authored by Bernhard Schölkopf seemed a good start.
Though I skimmed through the latter chapters, the beginning gives a good introduction to the different types of causality and which assumptions that have to be made. I especially liked the chapters drawing links between causality and topics like transfer learning and domain adaptation!
Though I skimmed through the latter chapters, the beginning gives a good introduction to the different types of causality and which assumptions that have to be made. I especially liked the chapters drawing links between causality and topics like transfer learning and domain adaptation!
March 22, 2021
Good intro to the modern Causality, and its usage for machine learning.
Interesting ideas:
- The language of Causal Graphs, and Structural Causal Models (SCM) are so powerful! They fix the problems that statistical models cannot do (e.g., interventions, counterfactual reasoning, etc.)
- To understand the world (composed of observations of N factors p(x1, x2, x3, ..., xN) ), the most succinct/transferrable/robust way is to interpret it by causal mechanisms, e.g., decomposing the joint distribution of N factors into \prod_i p(x_i | PA_i ). This factorization will also induce the least Kolmogorov complexity, which can be approximated by the minimum description length (Grunwald 2007) (my review)
- Do-calculus is another powerful language, based on the causal graphs. It turns observational data to interventional results.
More readings:
- The recent summary of causality by the authors (2021) "Towards Causal Representation Learning": https://arxiv.org/pdf/2102.11107.pdf
- My Github repo of causality papers: https://github.com/zhijing-jin/Causal...
Interesting ideas:
- The language of Causal Graphs, and Structural Causal Models (SCM) are so powerful! They fix the problems that statistical models cannot do (e.g., interventions, counterfactual reasoning, etc.)
- To understand the world (composed of observations of N factors p(x1, x2, x3, ..., xN) ), the most succinct/transferrable/robust way is to interpret it by causal mechanisms, e.g., decomposing the joint distribution of N factors into \prod_i p(x_i | PA_i ). This factorization will also induce the least Kolmogorov complexity, which can be approximated by the minimum description length (Grunwald 2007) (my review)
- Do-calculus is another powerful language, based on the causal graphs. It turns observational data to interventional results.
More readings:
- The recent summary of causality by the authors (2021) "Towards Causal Representation Learning": https://arxiv.org/pdf/2102.11107.pdf
- My Github repo of causality papers: https://github.com/zhijing-jin/Causal...
May 7, 2022
So far, it is very well-written: up-to-date citations referring to state-of-the-art modelling of causal inference, complete, insightful, and accessible to the newcomers introductions of every substantial topic of causal modelling.
However, *every single* page is not self-contained, with multiple explanations per page being substituted by references to other passages of the book (e.g., referring to (other) exercises, theorems definitions, proof techniques, etc). To make things worse, many of these passages belong to *future* chapters. E.g., instead of an explanation of certain steps of a theorem's proof appearing in chapter 3.2, there are references to theorems in chapter 6.5. At the same time, chapter 6.5 is based on chapter 3 - and other previous chapters. Hence, if I want to understand the theorem of chapter 6.5 so as to understand the theorem of chapter 3.2, I will have to read a considerable amount of (new) theory, from page 0 to chapter 6.5, including chapter 3.2, the chapter I want to understand.
However, *every single* page is not self-contained, with multiple explanations per page being substituted by references to other passages of the book (e.g., referring to (other) exercises, theorems definitions, proof techniques, etc). To make things worse, many of these passages belong to *future* chapters. E.g., instead of an explanation of certain steps of a theorem's proof appearing in chapter 3.2, there are references to theorems in chapter 6.5. At the same time, chapter 6.5 is based on chapter 3 - and other previous chapters. Hence, if I want to understand the theorem of chapter 6.5 so as to understand the theorem of chapter 3.2, I will have to read a considerable amount of (new) theory, from page 0 to chapter 6.5, including chapter 3.2, the chapter I want to understand.
July 10, 2020
Good. More like a giant survey paper than a textbook, but honestly that's what I want.
Update 10/07/2020: it's not an *ideal* textbook on causality, but it is far and away the best book on causality I've found. Unlike Pearl it gives a *reasonably* rigorous treatment of the field, and the authors are still quite active in causality (half the papers I read are from them or their academic children).
Update 10/07/2020: it's not an *ideal* textbook on causality, but it is far and away the best book on causality I've found. Unlike Pearl it gives a *reasonably* rigorous treatment of the field, and the authors are still quite active in causality (half the papers I read are from them or their academic children).
Displaying 1 - 5 of 5 reviews