We have hundreds of algos for ml, and all of them are fiddly with dozens of variables and other details to tweak. One of the first areas when venturing into ml is "you have to put in lots of analysis work".
I disagree.
Instead, I propose the following. Identify the problem at hand, and then select ml algos that you believe will solve it ideally. Use ml to tweak the variables for each algorithm to give best settings, then compare the top for ideal.
Why should I choose the algorithm when discussing ml when the machine can do it for me?
something like this would also be much appreciated with statistics, which has a lot of jargon. for example, I'm going though a stats inference course and am having a hard time keeping straight the difference between similar but different things such as: standard error, standard deviation, and sample standard deviation
So this book must be awesome and I also have been looking around to find more on model-based ML stuff to read.
I guess this is a part of the 2013 Microsoft Research[1] paper
---------------
[1] http://research.microsoft.com/en-us/um/people/cmbishop/downl...