A book I'm enjoying so far is Richard McElreath's
Statistical Rethinking.
Xcelab |
remove preview |
|
Statistical Rethinking |
"This is a rare and valuable book that combines readable explanations, computer code, and active learning." -Andrew Gelman, Columbia University "...an impressive book that I do not hesitate recommending for prospective data analysts and applied statisticians!" -Christian Robert, Université Paris-Dauphine ( review) "A pedagogical masterpiece..." |
View this on Xcelab > |
|
|
I have been learning about all kinds of Machine Learning, mostly guided by Christopher Bishop's
Pattern Recognition and Machine Learning, a wonderful, consistent presentation of diverse techniques. But I am especially interested in multilevel models as an alternative to the current tendency in ML practicioners to use ever more complex neural networks. Neural networks are great, and things like convolutional neural networks, long short-term memory, and reinforcement learning are achieving great results. Interpretability, though, for these models is even more cutting edge than the techniques themselves.
Whereas neural networks shine when interpretability is not the main goal, data is massively plentiful, and computing power is abundant, multilevel models shine when data is not too big to fit in the memory of a single machine, and that's pretty often. Do a web search for "you don't have big data" to read pieces by many authors reminding us that very often, useful and sufficient datasets can fit into the RAM of a single machine.
McElreath's
Statistical Rethinking emphasizes the relationship between a human mind engaged in scientific inquiry and the statistical tools that can help. It builds up concepts from first principles in very quick and convincing arguments. Those explanations will help me to inspire confidence in peers when I use the statistical tools described in the book. Finally, it uses R and Stan, two great open source software packages, to provide concrete examples of the statistical techniques the book covers.
The book makes the statement that multilevel modeling should be the default way of approaching statistical modeling, and it supports that thesis well. Other books have made the same claim, including Gelman and Hill's
Data Analysis Using Regression and Multilevel/Hierarchical Models, but McElreath has a uniquely effective way of bridging the natural world we live in and the statistical world we work in.
------------------------------
Edward Cashin
Research Scientist II
------------------------------
Original Message:
Sent: 09-26-2016 11:03
From: Lara Harmon
Subject: The Book Recs Thread
Hello, all!
The first book recommendations thread was such a success (more than 30 suggested books--take a look!) that I thought I'd make an official 'evergreen' Book Recs Thread. So here we go! Anyone can share a stats-related book they've read and loved here, at any time, and not have to worry that they might be resurrecting a dead conversation.
Oh, and feel free to recommend other types of media, as well! If you're lucky enough to have come across a film, television show, or other piece of pop culture that features statisticians, this is the place to share.
Happy reccing!
- Lara
------------------------------
Lara Harmon
Marketing and Online Community Coordinator
American Statistical Association
------------------------------