allow sumlargest(x,k), sumsmallest(x,k), and sumlargesteigs(x,k) for k=0 (#429)

* fix 423

* bump version, fix NEWS

Author: ericphanson

NEWS.md

@@ -1,7 +1,15 @@
-# Changes in v0.14.3
+# Changes in v0.14.5
+
+* allow `sumlargest(x,k)`, `sumsmallest(x,k)`, and `sumlargesteigs(x,k)` for `k=0` (simply returns `Constant(0)`). ([#429](https://github.com/jump-dev/Convex.jl/pull/429)).
+
+# Changes in v0.14.4
 
 * fixed a bug where the values of variables were being converted to `Float64` even if the problem was solved in high precision. ([#427](https://github.com/jump-dev/Convex.jl/pull/427)).
 
+# Changes in v0.14.3
+
+* update compatibility bounds for BenchmarkTools 0.6
+
 # Changes in v0.14.2
 
 * added lasso, ridge, and elastic net regression examples ([#420](https://github.com/jump-dev/Convex.jl/pull/420)). Thanks to @PaulSoderlind!

Project.toml

@@ -1,6 +1,6 @@
 name = "Convex"
 uuid = "f65535da-76fb-5f13-bab9-19810c17039a"
-version = "0.14.4"
+version = "0.14.5"
 
 [deps]
 AbstractTrees = "1520ce14-60c1-5f80-bbc7-55ef81b5835c"

src/atoms/lp_cone/sumlargest.jl

@@ -60,5 +60,5 @@ function conic_form!(x::SumLargestAtom, unique_conic_forms::UniqueConicForms)
     return get_conic_form(unique_conic_forms, x)
 end
 
-sumlargest(x::AbstractExpr, k::Int) = SumLargestAtom(x, k)
-sumsmallest(x::AbstractExpr, k::Int) = -SumLargestAtom(-x, k)
+sumlargest(x::AbstractExpr, k::Int) = k == 0 ? Constant(0) : SumLargestAtom(x, k)
+sumsmallest(x::AbstractExpr, k::Int) = k == 0 ? Constant(0) : -SumLargestAtom(-x, k)

src/atoms/sdp_cone/sumlargesteigs.jl

@@ -42,7 +42,7 @@ function evaluate(x::SumLargestEigs)
     eigvals(evaluate(x.children[1]))[end-x.children[2]:end]
 end
 
-sumlargesteigs(x::AbstractExpr, k::Int) = SumLargestEigs(x, Constant(k))
+sumlargesteigs(x::AbstractExpr, k::Int) = k == 0 ? Constant(0) : SumLargestEigs(x, Constant(k))
 
 # Create the equivalent conic problem:
 #   minimize sk + Tr(Z)

src/problem_depot/problems/lp.jl

@@ -147,7 +147,7 @@ end
 
 @add_problem lp function lp_sumlargest_atom(handle_problem!, ::Val{test}, atol, rtol, ::Type{T}) where {T, test}
     x = Variable(2)
-    p = minimize(sumlargest(x, 2), x >= [1; 1]; numeric_type = T)
+    p = minimize(sumlargest(x, 2) + sumlargest(x,0), x >= [1; 1]; numeric_type = T)
 
     if test
         @test vexity(p) == ConvexVexity()
@@ -173,7 +173,7 @@ end
 
 @add_problem lp function lp_sumsmallest_atom(handle_problem!, ::Val{test}, atol, rtol, ::Type{T}) where {T, test}
     x = Variable(4, 4)
-    p = minimize(sumlargest(x, 2), sumsmallest(x, 4) >= 1; numeric_type = T)
+    p = minimize(sumlargest(x, 2) + sumsmallest(x,0), sumsmallest(x, 4) >= 1; numeric_type = T)
 
     if test
         @test vexity(p) == ConvexVexity()

src/problem_depot/problems/sdp.jl

@@ -221,7 +221,7 @@ end
     end
 
     x = Semidefinite(3)
-    p = minimize(sumlargesteigs(x, 2), [x[i,:] >= i for i=1:3]...; numeric_type = T)
+    p = minimize(sumlargesteigs(x, 2) + sumlargesteigs(x,0), [x[i,:] >= i for i=1:3]...; numeric_type = T)
 
     handle_problem!(p)
 

test/test_utilities.jl

@@ -420,6 +420,10 @@ end
         c = 1.0
         @test evaluate(Constant(c)) ≈ c
         @test Constant(c).size == (1,1)
+
+        @test evaluate(sumlargesteigs(Variable(4, 4), 0)) == 0
+        @test evaluate(sumlargest(Variable(4), 0)) == 0
+        @test evaluate(sumsmallest(Variable(4), 0)) == 0
     end
 
     @testset "Base.vect" begin