Improved Probability Support

[Eugenio Bargiacchi]

Hello everyone,

Unfortunately I was not able to participate to the last meeting, and I'm still not yet sure I will be able to participate to the next. Thus I wanted to briefly mention here one of the specific topics that I am interested about, and see if it's of interest to anyone else.

I mainly work with discrete probabilities, and I'd love to see an improved C++ support for distributions and sampling.

For example, there is currently no easy way to obtain standard deviation and mean for all distributions shipped in random - along any other number of interesting functions such as pdf, cdf, mode, and so on. This makes it hard to write generic code that supports multiple distributions, unless all these functions are written explicitly.

I'd also love to have a lightweight wrapper alternative to std::discrete_distribution to sample from existing unit vectors without copying. As it is now, for me std::discrete_distribution is basically unusable, as I store my probabilities in Eigen matrices, and I'm not going to move them out or copy them.

Additionally, it would also be useful to compute statistics from ranges of values (maybe the ranges proposal already includes some stuff, I'm not sure), such as mean, median, std, variance, etc. This would make it very convenient to run experiments and extract useful information to plot.

Best,
Eugenio

[Michael Wong]

Hi Eugenio. I would too. One thing it occurs to me is we need a similar layer priority list of things to work on similar to the Linear Algebra approach. That is one of the discussion we should have tomorrow.

On a side note, is there an alternate time that you can make the call. The compromise schedule is not set in stone and is uncomfortable for most (for me too) and is just the least bad. There likely will have an opportunity to change on the next DST change over. So the earlier I am aware of this I can see how I can accomodate it.