From 1d467aff827d45852e4c28618b03304660ec890e Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Beno=C3=AEt=20Legat?= <benoit.legat@gmail.com>
Date: Tue, 23 Jan 2024 13:56:15 +0100
Subject: [PATCH] Remove trailing commas

---
 .../examples/sensitivity-analysis-ridge.jl    | 58 +++++++++++++------
 docs/src/examples/sensitivity-analysis-svm.jl |  2 +-
 2 files changed, 42 insertions(+), 18 deletions(-)

diff --git a/docs/src/examples/sensitivity-analysis-ridge.jl b/docs/src/examples/sensitivity-analysis-ridge.jl
index 0309fb59..dd523753 100644
--- a/docs/src/examples/sensitivity-analysis-ridge.jl
+++ b/docs/src/examples/sensitivity-analysis-ridge.jl
@@ -33,9 +33,9 @@
 using JuMP
 import DiffOpt
 import Random
-import Ipopt
+import SCS
 import Plots
-using LinearAlgebra: dot
+using LinearAlgebra: dot, norm
 
 # ## Define and solve the problem
 
@@ -59,14 +59,14 @@ Y = w * X .+ b + 0.8 * randn(N);
 function fit_ridge(X, Y, alpha = 0.1)
     N = length(Y)
     ## Initialize a JuMP Model with Ipopt solver
-    model = Model(() -> DiffOpt.diff_optimizer(Ipopt.Optimizer))
+    model = Model(() -> DiffOpt.diff_optimizer(SCS.Optimizer))
     set_silent(model)
     @variable(model, w) # angular coefficient
     @variable(model, b) # linear coefficient
     ## expression defining approximation error
     @expression(model, e[i = 1:N], Y[i] - w * X[i] - b)
     ## objective minimizing squared error and ridge penalty
-    @objective(model, Min, 1 / N * dot(e, e) + alpha * (w^2),)
+    @objective(model, Min, 1 / N * dot(e, e) + alpha * (w^2))
     optimize!(model)
     return model, w, b # return model & variables
 end
@@ -115,21 +115,45 @@ b̂ = value(b)
 
 # Sensitivity with respect to x and y
 
-∇y = zero(X)
-∇x = zero(X)
-for i in 1:N
-    MOI.set(
-        model,
-        DiffOpt.ForwardObjectiveFunction(),
-        2w^2 * X[i] + 2b * w - 2 * w * Y[i],
-    )
-    DiffOpt.forward_differentiate!(model)
-    ∇x[i] = MOI.get(model, DiffOpt.ForwardVariablePrimal(), w)
-    MOI.set(model, DiffOpt.ForwardObjectiveFunction(), (2Y[i] - 2b - 2w * X[i]))
-    DiffOpt.forward_differentiate!(model)
-    ∇y[i] = MOI.get(model, DiffOpt.ForwardVariablePrimal(), w)
+function sensitivities(model_constructor)
+    MOI.set(model, DiffOpt.ModelConstructor(), model_constructor)
+    ∇y = zero(X)
+    ∇x = zero(X)
+    for i in 1:N
+        MOI.set(
+            model,
+            DiffOpt.ForwardObjectiveFunction(),
+            2w^2 * X[i] + 2b * w - 2 * w * Y[i],
+        )
+        DiffOpt.forward_differentiate!(model)
+        ∇x[i] = MOI.get(model, DiffOpt.ForwardVariablePrimal(), w)
+        MOI.set(
+            model,
+            DiffOpt.ForwardObjectiveFunction(),
+            (2Y[i] - 2b - 2w * X[i]),
+        )
+        DiffOpt.forward_differentiate!(model)
+        ∇y[i] = MOI.get(model, DiffOpt.ForwardVariablePrimal(), w)
+    end
+    return ∇x, ∇y
 end
 
+# The sensitivities can either be obtained with the conic DiffOpt model
+
+∇x, ∇y = sensitivities(DiffOpt.ConicProgram.Model)
+
+# Or with the quadratic DiffOpt model
+
+∇x_quad, ∇y_quad = sensitivities(DiffOpt.QuadraticProgram.Model)
+
+# We can see that the tangent `∇x` obtained in both cases are close
+
+norm(∇x - ∇x_quad)
+
+# The same is true for the tangent obtained for `∇y`
+
+norm(∇y - ∇y_quad)
+
 # Visualize point sensitivities with respect to regression points.
 
 p = Plots.scatter(
diff --git a/docs/src/examples/sensitivity-analysis-svm.jl b/docs/src/examples/sensitivity-analysis-svm.jl
index f948cb43..d5af2810 100644
--- a/docs/src/examples/sensitivity-analysis-svm.jl
+++ b/docs/src/examples/sensitivity-analysis-svm.jl
@@ -64,7 +64,7 @@ MOI.set(model, MOI.Silent(), true)
 
 # Define the objective and solve
 
-@objective(model, Min, λ * LinearAlgebra.dot(w, w) + sum(ξ),)
+@objective(model, Min, λ * LinearAlgebra.dot(w, w) + sum(ξ))
 
 optimize!(model)