finish norm transition

benchmark/alternating_minimization.jl

@@ -22,11 +22,8 @@ function alternating_minimization(f, A, M, Y_init, k, MAX_ITERS)
     X = Variable(m, k)
     Y = Variable(k, n)
 
-    objective = (
-        norm(vec(M .* A) - vec(M .* (X * Y)), 2) +
-        γ1 * norm(vec(X), 2) +
-        γ2 * norm(vec(Y), 1)
-    )
+    objective =
+        (norm(M .* A - M .* (X * Y), 2) + γ1 * norm(X, 2) + γ2 * norm(Y, 1))
 
     constraints = [X * Y >= ϵ]
 

docs/examples_literate/general_examples/basic_usage.jl

@@ -47,7 +47,7 @@ println(evaluate(x[1] + x[4] - x[2]))
 X = Variable(2, 2)
 y = Variable()
 ## X is a 2 x 2 variable, and y is scalar. X' + y promotes y to a 2 x 2 variable before adding them
-p = minimize(norm(vec(X)) + y, 2 * X <= 1, X' + y >= 1, X >= 0, y >= 0)
+p = minimize(norm(X) + y, 2 * X <= 1, X' + y >= 1, X >= 0, y >= 0)
 solve!(p, SCS.Optimizer; silent_solver = true)
 println(round.(evaluate(X), digits = 2))
 println(evaluate(y))

docs/examples_literate/supplemental_material/paper_examples.jl

@@ -35,7 +35,7 @@ X = Variable(m, n);
 A = randn(p, m);
 b = randn(p, n);
 @time begin
-    p = minimize(norm(vec(X)), A * X == b)
+    p = minimize(norm(X), A * X == b)
 end
 @time solve!(p, ECOS.Optimizer; silent_solver = true)
 
@@ -52,6 +52,6 @@ n = 3
 A = randn(n, n);
 #@time begin
 X = Variable(n, n);
-p = minimize(norm(vec(X' - A)), X[1, 1] == 1);
+p = minimize(norm(X' - A), X[1, 1] == 1);
 solve!(p, ECOS.Optimizer; silent_solver = true)
 #end

docs/src/release_notes.md

@@ -18,6 +18,7 @@ Other changes:
 * [Type piracy](https://docs.julialang.org/en/v1/manual/style-guide/#Avoid-type-piracy) of `imag` and `real` has been removed. This should not affect use of Convex. Unfortunately, piracy of `hcat`, `vcat`, and `hvcat` still remains.
 * `sumlargesteigs` now enforces that it's argument is hermitian.
 * Bugfix: `dot` now correctly complex-conjugates its first argument
+* `norm` on `AbstractExpr` objects now supports matrices (treating them like vectors), matching Base's behavior.
 
 ## v0.15.4 (October 24, 2023)
 

src/atoms/second_order_cone/norm.jl

@@ -8,39 +8,31 @@ norm_fro(x::AbstractExpr) = norm2(vec(x))
 """
     norm(x::AbstractExpr, p::Real=2)
 
-Computes the `p`-norm `‖x‖ₚ = (∑ᵢ |xᵢ|^p)^(1/p)` of a vector expression `x`.
+Computes the `p`-norm `‖x‖ₚ = (∑ᵢ |xᵢ|^p)^(1/p)` of a vector expression `x`. Matrices
+are vectorized (i.e. `norm(x)` is the same as `norm(vec(x))`.)
 
 This function uses specialized methods for `p=1, 2, Inf`. For `p > 1` otherwise,
 this function uses the procedure documented at
 [`rational_to_socp.pdf`](https://github.com/jump-dev/Convex.jl/raw/master/docs/supplementary/rational_to_socp.pdf),
 based on the paper "Second-order cone programming" by F. Alizadeh and D. Goldfarb,
 Mathematical Programming, Series B, 95:3-51, 2001.
 
-!!! warning
-    For versions of Convex.jl prior to v0.14.0, `norm` on a matrix expression returned
-    the operator norm ([`opnorm`](@ref)), which matches Julia v0.6 behavior. This functionality
-    was deprecated since Convex.jl v0.8.0, and has been removed. In the future,
-    `norm(x, p)` will return `‖vec(x)‖ₚ`, matching the behavior of [`norm`](@ref)
-    for numeric matrices.
 """
 function LinearAlgebra.norm(x::AbstractExpr, p::Real = 2)
-    if length(size(x)) <= 1 || minimum(size(x)) == 1
-        # x is a vector
-        if p == 1
-            return norm_1(x)
-        elseif p == 2
-            return norm2(x)
-        elseif p == Inf
-            return norm_inf(x)
-        elseif p > 1
-            # TODO: allow tolerance in the rationalize step
-            return rationalnorm(x, rationalize(Int, float(p)))
-        else
-            error("vector p-norms not defined for p < 1")
-        end
+    if size(x, 2) > 1
+        x = vec(x)
+    end
+    # x is a vector
+    if p == 1
+        return norm_1(x)
+    elseif p == 2
+        return norm2(x)
+    elseif p == Inf
+        return norm_inf(x)
+    elseif p > 1
+        # TODO: allow tolerance in the rationalize step
+        return rationalnorm(x, rationalize(Int, float(p)))
     else
-        error(
-            "In Convex.jl v0.13 and below, `norm(x, p)` meant `opnorm(x, p)` (but was deprecated since v0.8.0). In the future, `norm(x,p)` for matrices will be equivalent to `norm(vec(x),p)`. This is currently an error to ensure you update your code!",
-        )
+        error("vector p-norms not defined for p < 1")
     end
 end

src/problem_depot/problems/socp.jl

@@ -88,15 +88,15 @@ end
 ) where {T,test}
     m = Variable(4, 5)
     c = [m[3, 3] == 4, m >= 1]
-    p = minimize(norm(vec(m), 2), c; numeric_type = T)
+    p = minimize(norm(m, 2), c; numeric_type = T)
 
     if test
         @test problem_vexity(p) == ConvexVexity()
     end
     handle_problem!(p)
     if test
         @test p.optval ≈ sqrt(35) atol = atol rtol = rtol
-        @test evaluate(norm(vec(m), 2)) ≈ sqrt(35) atol = atol rtol = rtol
+        @test evaluate(norm(m, 2)) ≈ sqrt(35) atol = atol rtol = rtol
         @test p.constraints[1].dual ≈ 0.6761 atol = atol rtol = rtol
         dual = 0.1690 .* ones(4, 5)
         dual[3, 3] = 0
@@ -542,16 +542,11 @@ end
         @test evaluate(opnorm(x, Inf)) ≈ opnorm(A, Inf) atol = atol rtol = rtol
     end
     # Vector norm
-    # TODO: Once the deprecation for norm on matrices is removed, remove the `vec` calls
     if test
-        @test evaluate(norm(vec(x), 1)) ≈ norm(vec(A), 1) atol = atol rtol =
-            rtol
-        @test evaluate(norm(vec(x), 2)) ≈ norm(vec(A), 2) atol = atol rtol =
-            rtol
-        @test evaluate(norm(vec(x), 7)) ≈ norm(vec(A), 7) atol = atol rtol =
-            rtol
-        @test evaluate(norm(vec(x), Inf)) ≈ norm(vec(A), Inf) atol = atol rtol =
-            rtol
+        @test evaluate(norm(x, 1)) ≈ norm(vec(A), 1) atol = atol rtol = rtol
+        @test evaluate(norm(x, 2)) ≈ norm(vec(A), 2) atol = atol rtol = rtol
+        @test evaluate(norm(x, 7)) ≈ norm(vec(A), 7) atol = atol rtol = rtol
+        @test evaluate(norm(x, Inf)) ≈ norm(vec(A), Inf) atol = atol rtol = rtol
     end
 end