use abstol & reltol options to setup convergence metrics tolerance

Author: wildart

src/api/optimize.jl

@@ -77,6 +77,23 @@ function optimize(objfun::D, constraints::C, method::M, population::AbstractArra
     converged, counter_tol = false, 0 # tolerance convergence
     is_moo = ismultiobjective(objfun)
 
+    # initialize convergence metrics
+    # use abstol as a tolerance value for metrics
+    if !isinf(options.abstol)
+        for m in method.metrics
+            m.tol = options.abstol
+        end
+    end
+    # if reltol is set, try to add RelDiff metric
+    if !(isinf(options.reltol) || is_moo)
+        idx = findfirst(m->isa(m, RelDiff), method.metrics)
+        if idx === nothing
+            push!(method.metrics, RelDiff(options.reltol))
+        else
+            method.metrics[idx].tol = options.reltol
+        end
+    end
+
     options.show_trace && print_header(method)
     trace!(tr, iteration, objfun, state, population, method, options, time()-t0)
 

src/api/types.jl

@@ -92,8 +92,8 @@ const ConvergenceMetrics = Vector{ConvergenceMetric}
 # Options
 """
 There are following options available:
-- `abstol::Float64`: the absolute tolerance used in the convergence test (*deprecated, use `metrics` parameter of a particular optimization algorithm*)
-- `reltol::Float64`: the relative tolerance used in the convergence test (*deprecated, use `metrics` parameter of a particular optimization algorithm*)
+- `abstol::Float64`: the absolute tolerance used in the convergence test
+- `reltol::Float64`: the relative tolerance used in the convergence test
 - `successive_f_tol::Integer`: the additional number of the iterations of the optimization algorithm after the convergence test is satisfied (*default: 10*)
 - `iterations::Integer`: the total number of the iterations of the optimization algorithm (*default: 1000*)
 - `show_trace::Bool`: enable the trace information display during the optimization (*default: false*).

test/interface.jl

@@ -182,5 +182,18 @@
     @test Evolutionary.iterations(res) == 7
     @test !Evolutionary.converged(res)
     @test length(Evolutionary.trace(res)) > 1
+    @test length(res.metrics) == 1
+    @test res.metrics[1].tol == 1e-10
+
+    o = Evolutionary.Options(abstol=1e-3)
+    res =  Evolutionary.optimize(sum, (()->BitVector(rand(Bool,dimension))), mthd, o)
+    @test length(res.metrics) == 1
+    @test res.metrics[1].tol == 1e-3
+
+    o = Evolutionary.Options(abstol=1e-3, reltol=1e-4)
+    res =  Evolutionary.optimize(sum, (()->BitVector(rand(Bool,dimension))), mthd, o)
+    @test length(res.metrics) == 2
+    @test res.metrics[1].tol == 1e-3
+    @test res.metrics[2].tol == 1e-4
 
 end