dot() called on sparse matrices obeys 2D dimension matching rules,
even if they are single row/columns. dot() called on (1, N) and
(1, N) arguments, as was being done here, will fail. This transposes
the second argument to fix the problem.