Error in assignment of row vector to matrix row about stan HOT 5 CLOSED

I think it is a parser issue. Last I heard, it was not possible to assign a row vector to a row of a matrix. Also, when you declare a matrix, in C++ it is an Eigen::Matrix so [i] assigns to the i-th element in column-major order, rather than to the i-th row. And assigning anything to the i-th element other than a scalar shouldn't work.

from stan.

A matrix row/column is not a valid L-value in the modelling language, as the transformed C++ for row/column extraction produces a copy of the data, as opposed to a reference to it. Ideally this should be caught by the parser right now. I'm not sure hard that would be to implement.

from stan.

I didn't think that I could do it, but I had forgot to change a few lines of code before compiling. If the row_vectors are changed to vectors, then the parser catches the error.

INVALID COMMAND-LINE ARGUMENT
EXPECTATION FAILURE LOCATION: file=foo.stan; line=11, column=3

  alpha_beta[1] <- alpha;
  ^-- here


DIAGNOSTIC(S) FROM PARSER:
base type mismatch in assignment; left variable=alpha_beta; left base type=row vector; right base type=vector
binary infix operator <= with functional interpretation logical_lt requires arguments or primitive type (int or real), found left type=row vector, right arg type=vector; no matches for function name="logical_lt"
    arg 0 type=row vector
    arg 1 type=vector
Parser expecting: "}"

from stan.

This is definitely a bug. My intention (which was reflected
in the doc and the parser, but not the assignment implementation)
was for the following to be valid:

matass.stan

transformed data {
matrix[2,2] X;
row_vector[2] y;

X[1] <- y;
X[2] <- y;
print("X=",X);
}
model {

}

Right now, during compilation, the issue is this:

/Users/carp/temp2/matass.cpp:48:9: error: no matching function for call to 'assign'
assign(get_base1(X,1,"X",1), y);

I'll patch the code so that (a) get_base1(...) returns a reference
to a view of a row of X, and (b) assign does the right thing with
it.

• Bob

On 2/12/13 2:25 PM, Marcus Brubaker wrote:

A matrix row/column is not a valid L-value in the modelling language, as the transformed C++ for row/column extraction
produces a copy of the data, as opposed to a reference to it. Ideally this should be caught by the parser right now. I'm
not sure hard that would be to implement.

—
Reply to this email directly or view it on GitHub #12 (comment).

from stan.

[master 034ac1c] resolved issue #12: allowing matrix row assignments on LHS with vars

from stan.