Missing signatures for _jacobian and _constrain functions?

General
1

I was surprised to find that it is not possible to use the _jacobian and _constrain functions to get bounded vectors that share bounds. This

code <- "
  data {}
  parameters {vector[2] x_raw;}
  transformed parameters {vector[2] x = lower_upper_bound_jacobian(x_raw, -1.0, 1.0);}
  model {x ~ std_normal();}
"
cmdstanr::cmdstan_model(cmdstanr::write_stan_file(code))

fails with

Semantic error in '/tmp/Rtmpby1fqc/model-2a279857adbaa1.stan', line 4, column 40 to column 85:
   -------------------------------------------------
     2:    data {}
     3:    parameters {vector[2] x_raw;}
     4:    transformed parameters {vector[2] x = lower_upper_bound_constrain(x_raw, -1.0, 1.0);}
                                                 ^
     5:    model {x ~ std_normal();}
     6:  
   -------------------------------------------------

Ill-typed arguments supplied to function "lower_upper_bound_constrain":
(vector, real, real)
Available signatures:
(vector, real, vector) => vector
  The third argument must be vector but got real
(vector, vector, real) => vector
  The second argument must be vector but got real
(vector, vector, vector) => vector
  The second argument must be vector but got real
(row_vector, real, row_vector) => row_vector
  The first argument must be row_vector but got vector
(row_vector, row_vector, real) => row_vector
  The first argument must be row_vector but got vector
(Additional signatures omitted)

make: *** [make/program:66: /tmp/Rtmpby1fqc/model-2a279857adbaa1.hpp] Error 1

Is there a reason that the (vector, real, real) => vector version does not exist? Obviously one can always build the necessary vector but there should be no need.

2

I’ll have to check but I think this might just be an oversight by me. Thanks for pointing it out

1 Like
3

This was a silly mistake by me, will be fixed in Add missing (container, real, real) overloads for contraint functions by WardBrian · Pull Request #1589 · stan-dev/stanc3 · GitHub (sorry you have to wait a whole release!)

2 Likes
4

Sweet, thank you Brian.