Almost done!
@Ronis_BR I did hit another blocker today. CUDA.jl also needs an update after DataFrames.jl.
I am actually surprised that CUDA.jl depends on DataFrames.jl and PrettyTables.jl. There is definitely room for improvement there.
Opened an issue:
The direct calls to PrettyTables.jl in CUDA.jl are here:
It should be really simple to update! However, I cannot test since I have no nvidia gpu
I believe their CI infrastructure covers it. We can submit a PR and wait for action.
Awesome! Can you do this please while I am working in DataFrames.jl?
I don’t need any .pdf documents (I don’t quite understand the LLM suggestion from @karei here); just an indication of what has changed. The “everything might have changed” is quite daunting honestly. Or, if that’s easier: what has stayed the same?
@Ronis_BR we can only run CUDA.jl tests after DataFrames.jl is released with support to PrettyTables.jl v3. I can try to prepare the PR to CUDA.jl after your DataFrames.jl PR is ready
You probably can have all those answers here:
@juliohm The PR to migrate DataFrames.jl is already submitted:
Thanks for adding the additional flexibility with first_line_column_label
. Unfortunately, this broke my code, which was working with PrettyTables.jl v3.0.0. I use the LaTeX backend and don’t want to have any style for the first line. In PrettyTables.jl v3.0.0 I just passed LatexTableStyle(first_line_column_label=[])
, which did the job, but I think with v3.0.2 this doesn’t work anymore.
ERROR: MethodError: no method matching LatexTableStyle(::Vector{…}, ::Vector{…}, ::Vector{…}, ::Vector{…}, ::Vector{…}, ::Vector{…}, ::Vector{…}, ::Vector{…}, ::Vector{…}, ::Vector{…}, ::Vector{…}, ::Vector{…}, ::Vector{…}, ::Vector{…}, ::Vector{…}, ::Vector{…})
The type `LatexTableStyle` exists, but no method is defined for this combination of argument types when trying to construct it.
How can I remove the default bold style for the first line with the latest version of PrettyTables.jl (in the LaTeX backend, but probably this is the same for the other backends)?
Edit: I think I found out: first_line_column_label=String[]
does it.
Yes! One significant change I made was to limit the number of types allowed within the structures. This modification resulted in a substantial performance boost. However, the “downside” is that we must ensure the correct types are provided, and []
is represented as Vector{Any}
.
@Ronis_BR how to replace the old crop=:none
option?
Hi @juliohm
In this case, you should use fit_table_in_display_horizontally = false, fit_table_in_display_vertically = false
.
Thank you @Ronis_BR , one more package migrated to PrettyTables.jl v3
I am traveling without access to my workstation, so I can’t easily prepare a PR to CUDA.jl until September 15. Please feel free to step in if you have a quick fix.
Thansk for your work! I found version 2 very well done and useful and I am sure that this update will be even better.
@Ronis_BR CUDA.jl requires non-trivial changes to update to PrettyTables.jl v3. Could you please give some hand over there?
Attempted a PR:
Hi @juliohm !
I do not have access to any computer with a Nvidia video card to test Any idea how can I test the modifications?
I have a NVIDIA GPU on my workstation, but the CUDA.jl tests crash for me in the master branch. Julia is killed to avoid memory shortage.
I also have no idea how the CUDA profiler should be tested.
cc: @maleadt
Did you try:
using Pkg
Pkg.test("CUDA",test_args=`--jobs=2`)
?
The default number of jobs might be too high for the memory you have.