What is Vector{TF} type?

[tjk9501]

I am using the projection function with TV, I thought a Float32. type m will work, but the follow error prompted:

julia> m_prj = prj(m_prj)
ERROR: MethodError: no method matching project_l1_Duchi!(::Vector{Float32}, ::Vector{Float32})
Closest candidates are:
  project_l1_Duchi!(::Union{Vector{TF}, Array{Complex{TF}, 1}}, ::TF) where TF<:Real at /home/hp/.julia/packages/SetIntersectionProjection/2oqD4/src/projectors/project_l1_Duchi!.jl:21
Stacktrace:
 [1] (::SetIntersectionProjection.var"#176#200"{set_definitions})(x::Vector{Float32})
   @ SetIntersectionProjection ~/.julia/packages/SetIntersectionProjection/2oqD4/src/get_projector.jl:33
 [2] PARSDMM_initialize(x::Vector{Float32}, l::Vector{Any}, y::Vector{Any}, AtA::Vector{Matrix{Float32}}, TD_OP::Vector{Union{JOLI.joAbstractLinearOperator{Float32, Float32}, SparseArrays.SparseMatrixCSC{Float32, Int64}}}, set_Prop::set_properties, P_sub::Vector{Any}, comp_grid::JUDI.IsoModel{Float32, 2}, maxit::Int64, rho_ini::Vector{Real}, gamma_ini::Float32, x_min_solver::String, rho_update_frequency::Int64, adjust_gamma::Bool, adjust_rho::Bool, adjust_feasibility_rho::Bool, m::Vector{Float32}, parallel::Bool, options::PARSDMM_options, zero_ini_guess::Bool, feasibility_only::Bool)
   @ SetIntersectionProjection ~/.julia/packages/SetIntersectionProjection/2oqD4/src/PARSDMM_initialize.jl:98
 [3] macro expansion
   @ ~/.julia/packages/SetIntersectionProjection/2oqD4/src/PARSDMM.jl:58 [inlined]
 [4] macro expansion
   @ ~/.julia/packages/TimerOutputs/RsWnF/src/TimerOutput.jl:237 [inlined]
 [5] PARSDMM(m::Vector{Float32}, AtA::Vector{Matrix{Float32}}, TD_OP::Vector{Union{JOLI.joAbstractLinearOperator{Float32, Float32}, SparseArrays.SparseMatrixCSC{Float32, Int64}}}, set_Prop::set_properties, P_sub::Vector{Any}, comp_grid::JUDI.IsoModel{Float32, 2}, options::PARSDMM_options, x::Vector{Float32}, l::Vector{Any}, y::Vector{Any})
   @ SetIntersectionProjection ~/.julia/packages/SetIntersectionProjection/2oqD4/src/PARSDMM.jl:40
 [6] PARSDMM(m::Vector{Float32}, AtA::Vector{Matrix{Float32}}, TD_OP::Vector{Union{JOLI.joAbstractLinearOperator{Float32, Float32}, SparseArrays.SparseMatrixCSC{Float32, Int64}}}, set_Prop::set_properties, P_sub::Vector{Any}, comp_grid::JUDI.IsoModel{Float32, 2}, options::PARSDMM_options)
   @ SetIntersectionProjection ~/.julia/packages/SetIntersectionProjection/2oqD4/src/PARSDMM.jl:38
 [7] (::var"#3#4")(x::Vector{Float32})
   @ Main ./REPL[76]:1
 [8] prj(input::Matrix{Float32})
   @ Main ./REPL[89]:5
 [9] top-level scope
   @ REPL[90]:1

What is Vector{TF} type in julia and how to define it or convert Float32. to Vector{TF} ?

[mloubout]

Closest candidates are:
  project_l1_Duchi!(::Union{Vector{TF}, Array{Complex{TF}, 1}}, ::TF) where TF<:Real 

THe error message is saying what it expect, i.e a TF* *where* TF<:Real so in your case you clearly haveTF=Float32` but looks like somewhere you are providing a vector instead of a number

[tjk9501]

In addition to that, I have a second question. The Judi Package have parallization in forward modeling. In gradient calculating process, for example, if I use the JudiJacobian method

J=JudiJacobian(F,q)

and

g=J(d_pred-d_obs)

Are there any method to accelerating this step (not using other functions like lsqr)? Because in my case this step is very slow in iteration.

[mloubout]

J cost ab out 2X forward since it computes two PDEs so it's expected to be slower yes

[tjk9501]

So, this is my understanding, the addprocs() function is accelerating both forward modeling and Jacobian, but Jacobian is solving more PDEs so it is slower. We don't need to further find other ways to speed it up because that is all what we can do.

[mloubout]

JUDI is quite optimized, thanks to devito as a backend, so with parallel over sources within JUDI there isn't much more that can be done yes.