set_vars! function to easily set variables in PrognosticVariables #154

maximilian-gelbrecht · 2022-09-27T14:08:59Z

This is a currently a draft, I want to get a review on it before I fully finish it.

I implemented set....!(prong,...) functions for all variables for spectral, grid and matrix input. These inputs are currently all Vector{...}
no doc strings yet
initializiation routines could be reworked with this as well
maybe there are some other parts of the code that could use this @milankl ?

I was also thinking of implementing the opposing get....(progn...) functions that return the Vectors{...}. What do you think?

A small observation: for the ModelSetup there should probably be a has(M::ModelSetup, var_name::Symbol) function, that could make a small minor things a bit more transparent.

milankl · 2022-09-27T16:10:10Z

* I implemented `set....!(prong,...)` functions for all variables for spectral, grid and matrix input. These inputs are currently all `Vector{...}`

Maybe we could even add methods like set_vorticity!(progn,L,lf=1) such that for a L being a LowerTriangularMatrix, or Matrix, or Gridded, (and not a vector of it) one can easily set the right leapfrog time step lf or layer layer etc.

* initializiation routines could be reworked with this as well
* maybe there are some other parts of the code that could use this @milankl ?

I think only the initial conditions actually set_...! the prognostic variables.

I was also thinking of implementing the opposing get....(progn...) functions that return the Vectors{...}. What do you think?

Yes, the time integration has lines like

vor_old  = progn.leapfrog[1].vor
vor_new  = progn.leapfrog[2].vor

which could be wrapped more generally into vor_old, vor_new = get_vorticity(progn,lf=(1,2)) but maybe it's not worth it?

A small observation: for the ModelSetup there should probably be a has(M::ModelSetup, var_name::Symbol) function, that could make a small minor things a bit more transparent.

Yes, at the moment, we set up all prognostic variables as a dummy 0-element array when they are not used in a given ModelSetup. E.g. in the BaroptropicModel

julia> p.layers[1].leapfrog[1].div
0×0 LowerTriangularMatrix{ComplexF64}

So if it's useful for you please implement has(M::ModelSetup, var_name::Symbol). At the moment we have lines like

M isa BarotropicModel ? nothing : gridded!(diagn.surface.pres_grid,progn.pres.leapfrog[lf],S)

which could then take the form like

has(M,:pres) ? nothing : gridded!(diagn.surface.pres_grid,progn.pres.leapfrog[lf],S)

I find both readable, but your suggestions externalises that property of M to a has function which we can define in models.jl to allow them to be more easily extended? I like that.

src/prognostic_variables.jl

milankl · 2022-09-27T16:16:45Z

src/prognostic_variables.jl

+                  var::Vector{<:LowerTriangularMatrix},
+                  lf::Integer=1) where NF


If these lines are changed to

var::LowerTriangularMatrix...; lf::Integer=1) where NF

then both L and (L1,L2) could be used as input? (as ... just unravels the tuple?) Is that helpful or would that just complicate things further down?

src/prognostic_variables.jl

maximilian-gelbrecht · 2022-09-28T12:16:26Z

I would not add functions/functionality to set individual layers. It does make the functions more complicated (or needs a separate set of functions). I don't see a super important purpose for it, and with the get... functions it is also very easy to emulate that behaviour

vor = get_vorticity(progn)
vor[3] = ...
set_vorticity!(progn, vor)

maximilian-gelbrecht · 2022-09-28T14:35:01Z

From my side this is finished.
One important change: I made PrognosticVariables parametric on the ModelSetup. If no ModelSetup is given, then it defaults to the abstract type. I've done this for the has function, so that one can easily return which variables are stored within an instance of PrognosticVariables. Maybe this could also be useful for other stuff in the future.

milankl · 2022-09-28T15:38:17Z

src/initial_conditions.jl

+    for var_name in var_names
+        if has(progn_new, var_name) 
+            var = get_var(progn_old, var_name) 
+            set_var!(progn_new, var_name, var)


Neat. I like it.

milankl · 2022-09-28T15:42:07Z

src/models.jl

@@ -20,6 +20,8 @@ struct BarotropicModel{NF<:AbstractFloat, D<:AbstractDevice} <: ModelSetup{D}
    device_setup::DeviceSetup{D}
 end

+has(::Type{BarotropicModel{NF,D}}, var_name::Symbol) where {NF,D} = var_name in [:vor]


I believe the ,D} can be dropped as it's not used (I had to remind myself what D stands for again...)? As far as I understand multi-parametric types in Julia they work like type{T1,T2,T3,T4} to be like type{T1}{T2}{T3}

julia> struct ABC{A,B,C} a::Int b::Int end julia> abc = ABC{Float64,Float32,Float16}(1,2) ABC{Float64, Float32, Float16}(1, 2) julia> t(abc::ABC{T}) where T = T(abc.a) julia> t(abc) 1.0

So dispatch is only performed following the first type if the others are not specified.

I don't understand why exactly, but it is required, otherwise it won't work. I did try it before. Just again:

MethodError: no method matching has(::Type{SpeedyWeather.PrimitiveEquationModel{Float64, SpeedyWeather.CPUDevice}}, ::Symbol) Closest candidates are: has(::Type{SpeedyWeather.BarotropicModel{NF}}, ::Symbol) where NF at ~/Nextcloud/SpeedyFork/SpeedyWeather.jl/src/models.jl:23 has(::Type{SpeedyWeather.ShallowWaterModel{NF}}, ::Symbol) where NF at ~/Nextcloud/SpeedyFork/SpeedyWeather.jl/src/models.jl:46 has(::Type{SpeedyWeather.PrimitiveEquationModel{NF}}, ::Symbol) where NF at ~/Nextcloud/SpeedyFork/SpeedyWeather.jl/src/models.jl:71

src/models.jl

src/prognostic_variables.jl

milankl · 2022-09-28T15:49:51Z

src/prognostic_variables.jl

+    @assert length(var) == length(progn.layers)
+    @assert has(progn, varname)


Reminder to self that we should do something about error messages in general. I like the @assert here as they actually print what's wrong. However, often I ended up using throw(BoundsError) which is kind of clear if there's only one @boundscheck before a loop in a function but if there's several ones, you don't know which array sizes don't match. Maybe we can make them somehow more telling...

I didn't know that before, but @assert let's you specify error messages:
@assert has(progn, varname) "No variable with this name found."

milankl · 2022-09-28T16:02:47Z

Last question, while we are at it: Should we also include set_vorticity!(progn,0) to just set the vorticity to zero (or any other scalar? This could be done via

function set_var!(progn::PrognosticVariables{NF}, 
                  varname::Symbol, 
                  s::Number;
                  lf::Integer=1) where NF

    for progn_layer in zip(progn.layers
        fill!(getfield(progn_layer.leapfrog[lf], varname), s)
    end 

    return progn 
end

functions like set_vorticity! should just pass on s::Number etc?

maximilian-gelbrecht · 2022-09-29T07:38:49Z

Last question, while we are at it: Should we also include set_vorticity!(progn,0) to just set the vorticity to zero (or any other scalar? This could be done via
function set_var!(progn::PrognosticVariables{NF}, 
                  varname::Symbol, 
                  s::Number;
                  lf::Integer=1) where NF

    for progn_layer in zip(progn.layers
        fill!(getfield(progn_layer.leapfrog[lf], varname), s)
    end 

    return progn 
end 
functions like set_vorticity! should just pass on s::Number etc?

Yeah, why not, I'll add it

initial var_args commit

9486640

maximilian-gelbrecht added the array types 🌐 LowerTriangularMatrices and RingGrids label Sep 27, 2022

maximilian-gelbrecht requested a review from milankl September 27, 2022 14:09

maximilian-gelbrecht changed the title ~~initial var_args commit~~ Set_vars function to easily set variables in PrognosticVariables Sep 27, 2022

maximilian-gelbrecht changed the title ~~Set_vars function to easily set variables in PrognosticVariables~~ set_vars! function to easily set variables in PrognosticVariables Sep 27, 2022

maximilian-gelbrecht mentioned this pull request Sep 27, 2022

Vorticity initial conditions #153

Closed

milankl reviewed Sep 27, 2022

View reviewed changes

maximilian-gelbrecht added 2 commits September 28, 2022 15:48

update, get_... and has funcs

c6900ca

doc strings and initialize_from_file

49ad052

bug fix in initialize_from_file!

f96ab02

maximilian-gelbrecht requested a review from milankl September 28, 2022 15:15

forgot about pressure in the initialize_from_file!

a52cbc6

milankl reviewed Sep 28, 2022

View reviewed changes

docs strings and set_var! for scalar

6660f07

maximilian-gelbrecht merged commit 6f8e186 into SpeedyWeather:main Sep 29, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

set_vars! function to easily set variables in PrognosticVariables #154

set_vars! function to easily set variables in PrognosticVariables #154

maximilian-gelbrecht commented Sep 27, 2022 •

edited

milankl commented Sep 27, 2022

milankl Sep 27, 2022

maximilian-gelbrecht commented Sep 28, 2022

maximilian-gelbrecht commented Sep 28, 2022

milankl Sep 28, 2022

milankl Sep 28, 2022

milankl Sep 28, 2022

maximilian-gelbrecht Sep 29, 2022

milankl Sep 28, 2022

maximilian-gelbrecht Sep 29, 2022

milankl commented Sep 28, 2022

maximilian-gelbrecht commented Sep 29, 2022

		var::Vector{<:LowerTriangularMatrix},
		lf::Integer=1) where NF

		@assert length(var) == length(progn.layers)
		@assert has(progn, varname)

set_vars! function to easily set variables in PrognosticVariables #154

set_vars! function to easily set variables in PrognosticVariables #154

Conversation

maximilian-gelbrecht commented Sep 27, 2022 • edited

milankl commented Sep 27, 2022

milankl Sep 27, 2022

Choose a reason for hiding this comment

maximilian-gelbrecht commented Sep 28, 2022

maximilian-gelbrecht commented Sep 28, 2022

milankl Sep 28, 2022

Choose a reason for hiding this comment

milankl Sep 28, 2022

Choose a reason for hiding this comment

milankl Sep 28, 2022

Choose a reason for hiding this comment

maximilian-gelbrecht Sep 29, 2022

Choose a reason for hiding this comment

milankl Sep 28, 2022

Choose a reason for hiding this comment

maximilian-gelbrecht Sep 29, 2022

Choose a reason for hiding this comment

milankl commented Sep 28, 2022

maximilian-gelbrecht commented Sep 29, 2022

maximilian-gelbrecht commented Sep 27, 2022 •

edited