Stop overloading Base.minimum and Base.maximum

docs/src/user/minimization.md

@@ -176,7 +176,7 @@ julia> Optim.minimizer(res)
  -0.3333
  -1.49994
 
-julia> minimum(res)
+julia> Optim.minimum(res)
  -2.8333333205768865
 ```
 

ext/OptimMOIExt.jl

@@ -421,7 +421,7 @@ MOI.get(::Optimizer, ::MOI.DualStatus) = MOI.NO_SOLUTION
 
 function MOI.get(model::Optimizer, attr::MOI.ObjectiveValue)
     MOI.check_result_index_bounds(model, attr)
-    val = minimum(model.results)
+    val = Optim.minimum(model.results)
     if model.sense == MOI.MAX_SENSE
         val = -val
     end

src/Optim.jl

@@ -53,7 +53,6 @@ import StatsBase: var
 using LinearAlgebra:
     LinearAlgebra,
     Diagonal,
-    diag,
     Hermitian,
     Symmetric,
     rmul!,

src/api.jl

@@ -1,6 +1,6 @@
 Base.summary(io::IO, r::OptimizationResults) = summary(io, r.method) # might want to do more here than just return summary of the method used
 minimizer(r::OptimizationResults) = r.minimizer
-Base.minimum(r::OptimizationResults) = r.minimum
+minimum(r::OptimizationResults) = r.minimum
 iterations(r::OptimizationResults) = r.iterations
 iteration_limit_reached(r::OptimizationResults) = r.stopped_by.iterations
 trace(r::OptimizationResults) =

src/maximize.jl

@@ -80,7 +80,7 @@ function maximize(
 end
 
 maximizer(r::MaximizationWrapper) = minimizer(res(r))
-Base.maximum(r::MaximizationWrapper) = -minimum(res(r))
+maximum(r::MaximizationWrapper) = -minimum(res(r))
 Base.summary(io::IO, r::MaximizationWrapper) = summary(io, res(r))
 
 for api_method in (

src/multivariate/solvers/constrained/fminbox.jl

@@ -560,7 +560,7 @@ function optimize(
             dfbox.obj.h_calls[1] = 0
         end
         copyto!(x, minimizer(results))
-        boxdist = min(minimum(x - l), minimum(u - x))
+        boxdist = Base.minimum(((xi, li, ui),) -> min(xi - li, ui - xi), zip(x, l, u)) # Base.minimum !== minimum
         if show_trace > 0
             println()
             println("Exiting inner optimizer with x = ", x)

src/multivariate/solvers/constrained/samin.jl

@@ -304,7 +304,7 @@ function optimize(
                 else
                     converge = 1
                     x_converged = true
-                    x_absΔ = maximum(bounds)
+                    x_absΔ = Base.maximum(bounds) # maximum !== Base.maximum
                 end
             else
                 f_converged = false

src/multivariate/solvers/first_order/ngmres.jl

@@ -355,7 +355,8 @@ function update_state!(
     Aview = view(state.A, 1:curw, 1:curw)
     bview = view(state.b, 1:curw)
     # The outer max is to avoid δ=0, which may occur if A=0, e.g. at numerical convergence
-    δ = method.ϵ0 * max(maximum(diag(Aview)), method.ϵ0)
+    # Note that Base.maximum !== maximum
+    δ = method.ϵ0 * max(Base.maximum(@view(Aview[diagind(Aview)])), method.ϵ0)
     try
         α .= (Aview + δ * I) \ bview
     catch e

src/multivariate/solvers/second_order/newton_trust_region.jl

@@ -52,7 +52,7 @@ work well without these safeguards when the Hessian is not positive definite.
 function initial_safeguards(H, gr, delta, lambda)
     # equations are on p. 560 of [MORESORENSEN]
     T = eltype(gr)
-    λS = maximum(-diag(H))
+    λS = -Base.minimum(@view(H[diagind(H)])) # Base.minimum !== minimum
     # they state on the first page that ||⋅|| is the Euclidean norm
     gr_norm = norm(gr)
     Hnorm = opnorm(H, 1)

src/multivariate/solvers/zeroth_order/particle_swarm.jl

@@ -110,7 +110,7 @@ function initial_state(
         end
     else
         # user did not define number of particles
-        n_particles = maximum([3, length(initial_x)])
+        n_particles = max(3, length(initial_x))
     end
     c1 = T(2)
     c2 = T(2)
@@ -193,7 +193,7 @@ function update_state!(f, state::ParticleSwarmState{T}, method::ParticleSwarm) w
 
     if state.iteration == 0
         copyto!(state.best_score, state.score)
-        f.F = minimum(state.score)
+        f.F = Base.minimum(state.score) # Base.minimum !== minimum
     end
     f.F = housekeeping!(
         state.score,

src/utilities/assess_convergence.jl

@@ -6,13 +6,13 @@ f_relchange(f_x::T, f_x_previous) where {T} = abs(f_x - f_x_previous) / abs(f_x)
 x_abschange(state) = x_abschange(state.x, state.x_previous)
 x_abschange(x, x_previous) = maxdiff(x, x_previous)
 x_relchange(state) = x_relchange(state.x, state.x_previous)
-x_relchange(x, x_previous) = maxdiff(x, x_previous) / maximum(abs, x)
+x_relchange(x, x_previous) = maxdiff(x, x_previous) / Base.maximum(abs, x) # Base.maximum !== maximum
 
 g_residual(d, state) = g_residual(d)
 g_residual(d, state::NelderMeadState) = state.nm_x
 g_residual(d::AbstractObjective) = g_residual(gradient(d))
 g_residual(d::NonDifferentiable) = convert(typeof(value(d)), NaN)
-g_residual(g) = maximum(abs, g)
+g_residual(g) = Base.maximum(abs, g) # Base.maximum !== maximum
 gradient_convergence_assessment(state::AbstractOptimizerState, d, options) =
     g_residual(gradient(d)) ≤ options.g_abstol
 gradient_convergence_assessment(state::ZerothOrderState, d, options) = false
@@ -52,7 +52,7 @@ function assess_convergence(
     if x_abschange(x, x_previous) ≤ x_abstol
         x_converged = true
     end
-    if x_abschange(x, x_previous) ≤ x_reltol * maximum(abs, x)
+    if x_abschange(x, x_previous) ≤ x_reltol * Base.maximum(abs, x) # Base.maximum !== maximum
         x_converged = true
     end
 

src/utilities/trace.jl

@@ -16,7 +16,7 @@ function common_trace!(
         dt["g(x)"] = copy(gradient(d))
         dt["Current step size"] = state.alpha
     end
-    g_norm = maximum(abs, gradient(d))
+    g_norm = Base.maximum(abs, gradient(d)) # Base.maximum !== maximum
     update!(
         tr,
         iteration,

test/general/api.jl

@@ -1,5 +1,9 @@
 # Test multivariate optimization
 @testset "Multivariate API" begin
+    # We do not overload `Base.minimum` and `Base.maximum`
+    @test Optim.minimum !== minimum
+    @test Optim.maximum !== maximum
+
     rosenbrock = MultivariateProblems.UnconstrainedProblems.examples["Rosenbrock"]
     f = MVP.objective(rosenbrock)
     g! = MVP.gradient(rosenbrock)