Comments on: Probability distributions and object oriented programming http://www.johndcook.com/blog/2009/01/23/probability-distributions-and-object-oriented-programming/ The blog of John D. Cook Sat, 31 Jul 2010 07:45:27 -0400 http://wordpress.org/?v=2.8.4 hourly 1 By: John http://www.johndcook.com/blog/2009/01/23/probability-distributions-and-object-oriented-programming/comment-page-1/#comment-13414 John Mon, 16 Feb 2009 14:57:55 +0000 http://www.johndcook.com/blog/?p=1339#comment-13414 What do a gamma and a chi squared distribution have in common? They share any behavior common to all continuous distributions, so it would make sense for both classes to derive from a class like <code>ContinuousDistribution</code>. But is there anything more specific they share? Not much. Both distributions have support on [0, ∞), but so does the log-normal distribution, for example. Any behavior related to parameters is going to be a problem, and behavior that doesn't depend on parameters may also not depend on distribution family at all. Obviously a <code>ChiSquareDistribution</code> class has a lot in common <em>internally</em> with a <code>GammaDistribution</code> class. The former could be a thin wrapper around an instance of the latter. But in terms of interfaces, the two classes don't have much in common except what they also share with a lot of other distributions. What do a gamma and a chi squared distribution have in common? They share any behavior common to all continuous distributions, so it would make sense for both classes to derive from a class like ContinuousDistribution. But is there anything more specific they share? Not much. Both distributions have support on [0, ∞), but so does the log-normal distribution, for example. Any behavior related to parameters is going to be a problem, and behavior that doesn’t depend on parameters may also not depend on distribution family at all.

Obviously a ChiSquareDistribution class has a lot in common internally with a GammaDistribution class. The former could be a thin wrapper around an instance of the latter. But in terms of interfaces, the two classes don’t have much in common except what they also share with a lot of other distributions.

]]> By: The Friar http://www.johndcook.com/blog/2009/01/23/probability-distributions-and-object-oriented-programming/comment-page-1/#comment-13411 The Friar Mon, 16 Feb 2009 14:24:28 +0000 http://www.johndcook.com/blog/?p=1339#comment-13411 Is there a reason not to have <em>ChiSquareDistribution</em> and <em>GammaDistribution</em> each inherit from some class like <em>BareBonesGammaDistribution</em>, with the parameter-setting methods written individually into the former pair but with methods for getting values written once for all in the latter? That is, isn't it just a matter of needing another level of inheritance hierarchy? Is there a reason not to have ChiSquareDistribution and GammaDistribution each inherit from some class like BareBonesGammaDistribution, with the parameter-setting methods written individually into the former pair but with methods for getting values written once for all in the latter? That is, isn’t it just a matter of needing another level of inheritance hierarchy?

]]>