Create a distribution by mapping the output of another one

[adamreichold]

This is useful if consumers are to be given an opaque type implementing the Distribution trait, but the output of the provided implementations needs additional post processing, e.g. to attach compile time units of measurement.

[vks]

Does this have an advantage over dist.sample_iter().map(...)?

[adamreichold]

Does this have an advantage over dist.sample_iter().map(...)?

My main use case is if a consumer needs a distribution producing a type that includes units of measurement, e.g. Distribution<uom::si::f64::Length> which I can create with distance_in_meters.map(|val| uom::si::f64::Length::new::<uom::si::length::meter>(val)) when distance_in_meters: Distribution<f64>.

If I changed the requirement of the consumer from Distribution<T> to Iterator<Item=T>, this would force me to keep the PRNG instance borrowed for the whole duration the consumer exists and it would be rather counter intuitive for someone reading the code as the fact the parameter represents a random variable is obscured. As I am using this for simulation models which need to be as easy to explain/understand as possible, these limitations would be significant issues for me.

[adamreichold]

this would force me to keep the PRNG instance borrowed for the whole duration the consumer exists

I am actually not sure whether I could fulfil this at all: I currently store the distributions with my model parameters while the PRNG state is part of the model state. The above would imply that I have to make these parameters part of the model state. Additionally, I would have to split the RNG streams for each distribution as I cannot mutably borrow a single one. And I would have to do this again for each agent as those are relocated between processes.

[dhardy]

I am not opposed to the idea. Code looks fine to me.

[dhardy]

Seems good to me. @vks can merge if he agrees.

[dhardy]

Eratta: please add a note to the changelog.

[adamreichold]

Eratta: please add a note to the changelog.

Done.

[adamreichold]

The CI failure looks related to packed_simd and I think unrelated to these changes. Is there anything I should do about this?

[dhardy]

No. Unstable features break sometimes (no surprise).