Sampling from a discrete probability distribution

问题:

I am new to C++ and extremely surprised by the lack of accessible, common probability manipulation tools (i.e. the lack of things in Boost and the standard library). I've done a lot of scientific programming in other languages, but the standard and/or ubiquitious third party add-ons always include a full range of probability tools. A friend billed up Boost to be the equivalent ubiquitous add-on for C++, but as I read the Boost documentation, even it seems to have a dearth of what I would consider extremely elementary built-ins.

I cannot find a built in that takes some sort of array of discrete probabilities and produces an index chosen according to those probabilities. I can of course write my own function for this, but I just wanted to check whether I am missing a standard way to do this.

Having to write my own functions at such a low-level is a bad thing, I feel, but I am writing a new simulation module for a larger project that is all in C++. My usual go-to tactic would be to write it in Python and link the Python to the C++, but because several other people are going to have to manage this code once I finish it, and none of them know Python, I think it would be more prudent to deliver it to them all in C++.

More generally, what do people do in C++ for things like sampling from standard distributions, in particular something as basic as a multi-variate normal distribution?

回答1:

^{(Moved from comment.)}

Did you look at Boost.Math.StatisticalDistributions? Specifically, its Discrete Probability Distributions?

Boost is not a library, it's a collection of libraries, so it can sometimes be difficult to find exactly what you're looking for – but that doesn't mean it isn't there. ;-]

回答2:

Perhaps I'm misunderstanding your intention, but it seems to me what you want is simply std::discrete_distribution.

回答3:

As mentioned, you'll want to look at boost/math/distributions and friends to meet your needs.

Here's a very good, detailed tutorial on getting these working for you in Boost. You may also want to throw your weight behind stan as well, which looks quite promising within this space.

回答4:

You should do less C++ bashing, and more question asking - we try to be helpful and respectful on SO. Questions like yours are often tagged as inflammatory.

Boost::math seems to provide exactly what you're looking for: https://www.quantnet.com/cplusplus-statistical-distributions-boost/ - I'm not 100% sure about how well it handles multi-variate distributions though (nor am I an expert on statistics).

Get it here: http://www.boost.org/doc/libs/1_49_0/libs/math/doc/html/index.html

回答5:

Boost's math libraries are pretty good for working with different distributions, but if you are only interested in sampling (as in the problem you mentioned in your post), then looking at the boost Random libraries might be more germane to your task. This link shows how to simulate rolling a weighted die, for example.