Re:Information Theory, Inference, and Learning Algorithms (2...
by oneisone on Friday March 18, 2016 @08:08AM
(#11311202)
I read a chapter of the book, it criticizes non Bayesian statistics too much. For example at discussing p-values. The p-value method is for getting sound results but not for interpreting the particular value obtained in one experiment, for example 5% is for being right 19 of 20 times, the numbers obtained in the experiments doesn't change this.
[ Parent
| on HN ]
Re:Information Theory, Inference, and Learning Algorithms (2...
by CuriouslyC on Friday March 18, 2016 @09:53AM
(#11311765)
The author is trying to present a probabalistic framework for reasoning about the world. The issue with p-values is that they have a narrow band of utility, and can't really contribute much to the process of inference as a whole; you can say no to a nearly infinite number of absurd propositions and still not be able to say anything insightful about a system. On top of that, p-values are clearly not intuitive, as demonstrated by their widespread misuse.