2

I wonder if there is any good and easy to understand book on reviewing on when should a particular statistical model (binomial, logistic regression, etc) be used?

My background is in statistics, but most of my work is developing MLE estimation algorithm using matrix calculus. I am now starting to focus more on data analysis, which comes in many different ways, and I don't have a good understanding which models are most proper.

For example, I know for binary observations I should use logistic regression, but I don't know under which situations I should use a non informative prior (e.g. Jeffreys' prior). Also, since there are so many choice of models, I don't know in a broad way which one to pick.

Thanks.

Nick Cox
  • 48,377
  • 8
  • 110
  • 156
Jack2019
  • 583
  • 4
  • 10
  • 1
    Have you checked this (aggregated) thread: http://stats.stackexchange.com/questions/12386/machine-learning-cookbook-reference-card-cheatsheet I think you'll find it quite helpful. – usεr11852 Jul 19 '13 at 15:46
  • I would consider non-parametric or semiparametric methods whenever possible. – semibruin Jul 19 '13 at 19:36

1 Answers1

2

Check out this flowchart for machine learning: http://peekaboo-vision.blogspot.ru/2013/01/machine-learning-cheat-sheet-for-scikit.html

tarteflambee
  • 56
  • 5