Add DiagonalQNModel and SpectralGradientModel #94
Conversation
Codecov ReportBase: 92.31% // Head: 92.40% // Increases project coverage by
Additional details and impacted files@@ Coverage Diff @@
## main #94 +/- ##
==========================================
+ Coverage 92.31% 92.40% +0.08%
==========================================
Files 6 6
Lines 859 869 +10
==========================================
+ Hits 793 803 +10
Misses 66 66
Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here. ☔ View full report at Codecov. |
|
There was a problem hiding this comment.
That looks great, thanks @geoffroyleconte for the new feature. I just made a couple of comments to try solve the issues raised, and improve doc and tests.
julia> @code_warntype hess_op(nlp, x)
MethodInstance for NLPModels.hess_op(::DiagonalQNModel{Float64, Vector{Float64}, SimpleNLPModel{Float64, Vector{Float64}, NLPModelMeta{Float64, Vector{Float64}}}, NLPModelMeta{Float64, Vector{Float64}}}, ::Vector{Float64})
from hess_op(nlp::QuasiNewtonModel, x::AbstractVector; kwargs...) in NLPModelsModifiers at C:\Users\Geoffroy Leconte\.julia\dev\NLPModelsModifiers\src\quasi-newton.jl:155
Arguments
#self#::Core.Const(NLPModels.hess_op)
nlp::DiagonalQNModel{Float64, Vector{Float64}, SimpleNLPModel{Float64, Vector{Float64}, NLPModelMeta{Float64, Vector{Float64}}}, NLPModelMeta{Float64, Vector{Float64}}}
x::Vector{Float64}
Body::LinearOperators.DiagonalQN
1 ─ %1 = Core.NamedTuple()::Core.Const(NamedTuple())
│ %2 = Base.pairs(%1)::Core.Const(Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}}())
│ %3 = NLPModelsModifiers.:(var"#hess_op#20")(%2, #self#, nlp, x)::LinearOperators.DiagonalQN
└── return %3(Body is in red, but not displayed here on github). |
|
With my last commit: julia> @code_warntype hess_op(nlp, x)
MethodInstance for NLPModels.hess_op(::DiagonalQNModel{Float64, Vector{Float64}, SimpleNLPModel{Float64, Vector{Float64}, NLPModelMeta{Float64, Vector{Float64}}}, NLPModelMeta{Float64, Vector{Float64}}, LinearOperators.DiagonalQN{Float64, Int64, Vector{Float64}, LinearOperators.var"#84#85"{Vector{Float64}}}}, ::Vector{Float64})
from hess_op(nlp::QuasiNewtonModel, x::AbstractVector; kwargs...) in NLPModelsModifiers at C:\Users\Geoffroy Leconte\.julia\dev\NLPModelsModifiers\src\quasi-newton.jl:155
Arguments
#self#::Core.Const(NLPModels.hess_op)
nlp::DiagonalQNModel{Float64, Vector{Float64}, SimpleNLPModel{Float64, Vector{Float64}, NLPModelMeta{Float64, Vector{Float64}}}, NLPModelMeta{Float64, Vector{Float64}}, LinearOperators.DiagonalQN{Float64, Int64, Vector{Float64}, LinearOperators.var"#84#85"{Vector{Float64}}}}
x::Vector{Float64}
Body::LinearOperators.DiagonalQN{Float64, Int64, Vector{Float64}, LinearOperators.var"#84#85"{Vector{Float64}}}
1 ─ %1 = Core.NamedTuple()::Core.Const(NamedTuple())
│ %2 = Base.pairs(%1)::Core.Const(Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}}())
│ %3 = NLPModelsModifiers.:(var"#hess_op#20")(%2, #self#, nlp, x)::LinearOperators.DiagonalQN{Float64, Int64, Vector{Float64}, LinearOperators.var"#84#85"{Vector{Float64}}}
└── return %3(no red). |
|
There was a problem hiding this comment.
Thanks @geoffroyleconte for the new models!
If you use these QuasiNewtonModel could you add later a short tutorial about it in the package documentation?
|
Sure. You mean improving the doc of each individual QN model or an example in the /docs (I can see that there is no tutorial page on this package but I can add one later)? |
|
Yes, a tutorial-like page for the QN models. We are starting from scratch here, so anything will be helpful :) |
2 participants
Add DiagonalQNModel and SpectralGradientModel based upon the operators in LinearOperators.
I dropped NLPModels 0.18 because there were failures when testing the package with the counters and the show functions (AmplNLReader tests fail because of this).