update algorithm verbosity and ci

.JuliaFormatter.toml

@@ -0,0 +1 @@
+style = "yas"

.github/workflows/ci.yml

@@ -10,7 +10,8 @@ jobs:
       fail-fast: false
       matrix:
         version:
-          - '1.4'
+          - '1.6'
+          - 'lts'
           - '1' # automatically expands to the latest stable 1.x release of Julia
         os:
           - ubuntu-latest

Project.toml

@@ -1,14 +1,14 @@
 name = "OptimKit"
 uuid = "77e91f04-9b3b-57a6-a776-40b61faaebe0"
 authors = ["Jutho Haegeman"]
-version = "0.3.2"
+version = "0.4"
 
 [deps]
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 Printf = "de0858da-6303-5e67-8744-51eddeeeb8d7"
 
 [compat]
-julia = "1"
+julia = "1.6"
 
 [extras]
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"

src/cg.jl

@@ -1,17 +1,14 @@
 abstract type CGFlavor end
 
-struct ConjugateGradient{F<:CGFlavor,T<:Real,L<:AbstractLineSearch} <: OptimizationAlgorithm
-    flavor::F
-    maxiter::Int
-    gradtol::T
-    linesearch::L
-    restart::Int
-    verbosity::Int
-end
-function ConjugateGradient(; flavor=HagerZhang(), maxiter=typemax(Int), gradtol::Real=1e-8,
-                           restart=typemax(Int), verbosity::Int=0,
-                           linesearch::AbstractLineSearch=HagerZhangLineSearch())
-    return ConjugateGradient(flavor, maxiter, gradtol, linesearch, restart, verbosity)
+@kwdef struct ConjugateGradient{F<:CGFlavor,T<:Real,L<:AbstractLineSearch} <:
+              OptimizationAlgorithm
+    flavor::F = HagerZhang()
+    maxiter::Int = typemax(Int)
+    gradtol::T = 1e-8
+    linesearch::L = HagerZhangLineSearch()
+    restart::Int = typemax(Int)
+    verbosity::Int = 1
+    ls_verbosity::Int = 10
 end
 
 function optimize(fg, x, alg::ConjugateGradient;
@@ -85,7 +82,7 @@ function optimize(fg, x, alg::ConjugateGradient;
         if normgrad <= alg.gradtol || numiter >= alg.maxiter
             break
         end
-        verbosity >= 2 &&
+        verbosity >= 3 &&
             @info @sprintf("CG: iter %4d: f = %.12f, ‖∇f‖ = %.4e, α = %.2e, β = %.2e, nfg = %d",
                            numiter, f, normgrad, α, β, nfg)
 
@@ -101,14 +98,14 @@ function optimize(fg, x, alg::ConjugateGradient;
         # increase α for next step
         α = 2 * α
     end
-    if verbosity > 0
-        if normgrad <= alg.gradtol
+    if normgrad <= alg.gradtol
+        verbosity >= 2 &&
             @info @sprintf("CG: converged after %d iterations: f = %.12f, ‖∇f‖ = %.4e",
                            numiter, f, normgrad)
-        else
+    else
+        verbosity >= 1 &&
             @warn @sprintf("CG: not converged to requested tol: f = %.12f, ‖∇f‖ = %.4e",
                            f, normgrad)
-        end
     end
     history = [fhistory normgradhistory]
     return x, f, g, numfg, history

src/gd.jl

@@ -1,14 +1,9 @@
-struct GradientDescent{T<:Real,L<:AbstractLineSearch} <: OptimizationAlgorithm
-    maxiter::Int
-    gradtol::T
-    linesearch::L
-    verbosity::Int
-end
-function GradientDescent(; maxiter=typemax(Int),
-                         gradtol::Real=1e-8,
-                         verbosity::Int=0,
-                         linesearch::AbstractLineSearch=HagerZhangLineSearch())
-    return GradientDescent(maxiter, gradtol, linesearch, verbosity)
+@kwdef struct GradientDescent{T<:Real,L<:AbstractLineSearch} <: OptimizationAlgorithm
+    maxiter::Int = typemax(Int)
+    gradtol::T = 1e-8
+    linesearch::L = HagerZhangLineSearch()
+    verbosity::Int = 1
+    ls_verbosity::Int = 1
 end
 
 function optimize(fg, x, alg::GradientDescent;
@@ -56,21 +51,21 @@ function optimize(fg, x, alg::GradientDescent;
         if normgrad <= alg.gradtol || numiter >= alg.maxiter
             break
         end
-        verbosity >= 2 &&
+        verbosity >= 3 &&
             @info @sprintf("GD: iter %4d: f = %.12f, ‖∇f‖ = %.4e, α = %.2e, nfg = %d",
                            numiter, f, normgrad, α, nfg)
 
         # increase α for next step
         α = 2 * α
     end
-    if verbosity > 0
-        if normgrad <= alg.gradtol
+    if normgrad <= alg.gradtol
+        verbosity >= 2 &&
             @info @sprintf("GD: converged after %d iterations: f = %.12f, ‖∇f‖ = %.4e",
                            numiter, f, normgrad)
-        else
+    else
+        verbosity >= 1 &&
             @warn @sprintf("GD: not converged to requested tol: f = %.12f, ‖∇f‖ = %.4e",
                            f, normgrad)
-        end
     end
     history = [fhistory normgradhistory]
     return x, f, g, numfg, history

src/lbfgs.jl

@@ -23,40 +23,16 @@ LBFGS optimization algorithm.
         linesearch::AbstractLineSearch = HagerZhangLineSearch())`: Construct an LBFGS object with the specified parameters.
 
 """
-struct LBFGS{T<:Real,L<:AbstractLineSearch} <: OptimizationAlgorithm
-    m::Int
-    maxiter::Int
-    gradtol::T
-    acceptfirst::Bool
-    linesearch::L
-    verbosity::Int
+@kwdef struct LBFGS{T<:Real,L<:AbstractLineSearch} <: OptimizationAlgorithm
+    m::Int = 8
+    maxiter::Int = typemax(Int)
+    gradtol::T = 1e-8
+    acceptfirst::Bool = true
+    linesearch::L = HagerZhangLineSearch()
+    verbosity::Int = 1
+    ls_verbosity::Int = 1
 end
 
-"""
-    LBFGS(m::Int = 8; maxiter = typemax(Int), gradtol::Real = 1e-8, acceptfirst::Bool = true,
-        verbosity::Int = 0,
-        linesearch::AbstractLineSearch = HagerZhangLineSearch())
-
-Construct an LBFGS object with the specified parameters.
-
-## Arguments
-- `m::Int = 8`: The number of previous iterations to store for the limited memory BFGS approximation.
-- `maxiter::Int = typemax(Int)`: The maximum number of iterations.
-- `gradtol::Real = 1e-8`: The tolerance for the norm of the gradient.
-- `acceptfirst::Bool = true`: Whether to accept the first step of the line search.
-- `verbosity::Int = 0`: The verbosity level.
-- `linesearch::AbstractLineSearch = HagerZhangLineSearch()`: The line search algorithm to use.
-
-## Returns
-- `LBFGS`: The LBFGS object.
-
-"""
-LBFGS(m::Int=8; maxiter=typemax(Int), gradtol::Real=1e-8, acceptfirst::Bool=true,
-verbosity::Int=0,
-linesearch::AbstractLineSearch=HagerZhangLineSearch()) = LBFGS(m, maxiter, gradtol,
-                                                               acceptfirst, linesearch,
-                                                               verbosity)
-
 function optimize(fg, x, alg::LBFGS;
                   precondition=_precondition, (finalize!)=_finalize!,
                   retract=_retract, inner=_inner, (transport!)=_transport!,
@@ -119,7 +95,7 @@ function optimize(fg, x, alg::LBFGS;
         if normgrad <= alg.gradtol || numiter >= alg.maxiter
             break
         end
-        verbosity >= 2 &&
+        verbosity >= 3 &&
             @info @sprintf("LBFGS: iter %4d: f = %.12f, ‖∇f‖ = %.4e, α = %.2e, m = %d, nfg = %d",
                            numiter, f, normgrad, α, length(H), nfg)
 
@@ -183,14 +159,14 @@ function optimize(fg, x, alg::LBFGS;
             push!(H, (scale!(s, 1 / norms), scale!(y, 1 / norms), ρ))
         end
     end
-    if verbosity > 0
-        if normgrad <= alg.gradtol
+    if normgrad <= alg.gradtol
+        verbosity >= 2 &&
             @info @sprintf("LBFGS: converged after %d iterations: f = %.12f, ‖∇f‖ = %.4e",
                            numiter, f, normgrad)
-        else
+    else
+        verbosity >= 1 &&
             @warn @sprintf("LBFGS: not converged to requested tol: f = %.12f, ‖∇f‖ = %.4e",
                            f, normgrad)
-        end
     end
     history = [fhistory normgradhistory]
     return x, f, g, numfg, history