distributions not vectorized for matrices about stan HOT 4 CLOSED

We discussed this a couple of times on stan-dev. It was held up due to
discussions about what to accept. For 1.2.1, I think anything should be
legal on the LHS as long as the parameters are scalars. But I thought we
also agreed that if a parameter was not a scalar its structure has to match
strictly with the LHS, which I was fine with.

On Thu, Mar 7, 2013 at 6:15 PM, Matt Hoffman [email protected]:

Ideally, I'd like to be able to write

parameters {
matrix[M, N] Z;
}
model {
Z ~ normal(0, 1);
}

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/23
.

from stan.

I agree with all of that, I think. It's pretty unambiguous what you should
do when a parameter is a scalar or the same shape as the LHS, and much less
unambiguous when it's neither of those things. Maybe someday we could get
fancier, but I think this would be a good first step.

Matt

On Thu, Mar 7, 2013 at 3:55 PM, bgoodri [email protected] wrote:

We discussed this a couple of times on stan-dev. It was held up due to
discussions about what to accept. For 1.2.1, I think anything should be
legal on the LHS as long as the parameters are scalars. But I thought we
also agreed that if a parameter was not a scalar its structure has to
match
strictly with the LHS, which I was fine with.

On Thu, Mar 7, 2013 at 6:15 PM, Matt Hoffman [email protected]:

Ideally, I'd like to be able to write

parameters {
matrix[M, N] Z;
}
model {
Z ~ normal(0, 1);
}

—
Reply to this email directly or view it on GitHub<
https://github.com/stan-dev/stan/issues/23>
.

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/23#issuecomment-14594556
.

from stan.

Let me just clarify what I think the plan is.

If you have a function f(x_1,...,x_N) where each
of x_1, ..., X_N is "scalar", then you can use
ints, reals, vectors, arrays up to 2D, or matrices
for any of the x_n.

Arguments of type real and int are allowed anywhere.
All of the non-real and non-int arguments must all
have the same shape.

Shape Variables

K real[K], vector[K], row_vector[K],
unit[K], ordered[K], simplex[K],

K x K cov_matrix[K], corr_matrix[K]

M x N matrix[M,N], real[M,N]

• Bob

from stan.

Moved to to-do list.

from stan.