Compact printing for Adjoint vectors, and Diagonal
#40722
Conversation
dkarrasch
added
display and printing
| @@ -136,6 +136,11 @@ end | |||
| function Base.replace_in_print_matrix(A::Diagonal,i::Integer,j::Integer,s::AbstractString) | |||
| i==j ? s : Base.replace_with_centered_mark(s) | |||
| end | |||
| function Base.show(io::IO, A::Diagonal) | |||
There was a problem hiding this comment.
Should we add similar methods to the other banded matrix types?
There was a problem hiding this comment.
Could do, but maybe not this PR?
For the more complicated ones, there are choices like whether to print SymTridiagonal([1,2,3], [4,5]) or rather SymTridiagonal([1 4 0; 4 2 5; 0 5 3]).
In this PR, only Diagonal is the only upper-case constructor printed.
Note also that not all Adjoint or Transpose objects are affected, only the simplest ones.
There was a problem hiding this comment.
Xref #55347 for Bidiagonal.
|
There are some doctest failures, but the other test failures are unrelated. If there are no objections, it would be good to fix and merge this. |
| @@ -302,6 +302,16 @@ function Base.showarg(io::IO, v::Transpose, toplevel) | |||
| toplevel && print(io, " with eltype ", eltype(v)) | |||
| return nothing | |||
| end | |||
| function Base.show(io::IO, v::Adjoint{<:Real, <:AbstractVector}) | |||
There was a problem hiding this comment.
Note that this omits vectors of complex numbers... because I thought that printing 0 + 3im when the element is in fact 0 - 3im here is worse than failing to note the container type:
julia> (transpose([1,2,3im]), adjoint([1,2,3im]))
(transpose(Complex{Int64}[1 + 0im, 2 + 0im, 0 + 3im]), Complex{Int64}[1 + 0im 2 + 0im 0 - 3im])It also doesn't change how e.g. an array of matrices is printed, for the same reason -- printing [1 2; 3 4] which isn't == the element seems confusing:
julia> ([[1 2; 3 4], [5 6; 7 8]]',)
(Adjoint{Int64, Matrix{Int64}}[[1 3; 2 4] [5 7; 6 8]],)
There was a problem hiding this comment.
Should we add tests for the complex case, making it explicit that this was a design choice and not an oversight?
There was a problem hiding this comment.
I think the restrictions to v::Adjoint{<:Real, <:AbstractVector} and v::Transpose{<:Number, <:AbstractVector} are fairly obviously deliberate.
For the complex case, it would not be crazy to print adjoint(conj([1 + 0im, 2 + 0im, 0 - 3im])) as this also resolves to the right object (since conj is eager). Maybe I'd rather not have explicit tests blocking that.
This changes the compact printing to preserve more information -- an adjoint vector is not quite a matrix, and Diagonal wastes a lot of space: ```julia julia> (Diagonal(1:4), [5,6,7]', transpose(8:10)) # before ([1 0 0 0; 0 2 0 0; 0 0 3 0; 0 0 0 4], [5 6 7], [8 9 10]) julia> (Diagonal(1:4), [5,6,7]', transpose(8:10)) # after (Diagonal(1:4), adjoint([5, 6, 7]), transpose(8:10)) ``` Would have been better to do at the same time as 1.6's other printing changes, I guess. --------- Co-authored-by: Jishnu Bhattacharya <[email protected]>
This changes the compact printing to preserve more information -- an adjoint vector is not quite a matrix, and Diagonal wastes a lot of space:
Would have been better to do at the same time as 1.6's other printing changes, I guess.