Merge pull request #475 from sathvikbhagavan/sb/svm

refactor(SurrogatesSVM): use SurrogatesBase and cleanup code

lib/SurrogatesSVM/Project.toml

@@ -5,16 +5,20 @@ version = "0.1.0"
 
 [deps]
 LIBSVM = "b1bec4e5-fd48-53fe-b0cb-9723c09d164b"
-Surrogates = "6fc51010-71bc-11e9-0e15-a3fcc6593c49"
+SurrogatesBase = "89f642e6-4179-4274-8202-c11f4bd9a72c"
 
 [compat]
 LIBSVM = "0.8"
-Surrogates = "6"
+SafeTestsets = "0.1"
+Surrogates = "6.9"
+SurrogatesBase = "1.1"
+Test = "1"
 julia = "1.10"
 
 [extras]
 SafeTestsets = "1bc83da4-3b8d-516f-aca4-4fe02f6d838f"
+Surrogates = "6fc51010-71bc-11e9-0e15-a3fcc6593c49"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [targets]
-test = ["SafeTestsets", "Test"]
+test = ["SafeTestsets", "Surrogates", "Test"]

lib/SurrogatesSVM/src/SurrogatesSVM.jl

@@ -1,79 +1,68 @@
 module SurrogatesSVM
 
-import Surrogates: AbstractSurrogate, add_point!
-export SVMSurrogate
-
+using SurrogatesBase
 using LIBSVM
 
-mutable struct SVMSurrogate{X, Y, M, L, U} <: AbstractSurrogate
+export SVMSurrogate, update!
+
+mutable struct SVMSurrogate{X, Y, M, L, U} <: AbstractDeterministicSurrogate
     x::X
     y::Y
     model::M
     lb::L
     ub::U
 end
 
-function SVMSurrogate(x, y, lb::Number, ub::Number)
-    xn = reshape(x, length(x), 1)
-    model = LIBSVM.fit!(SVC(), xn, y)
-    SVMSurrogate(xn, y, model, lb, ub)
-end
+"""
+    SVMSurrogate(x, y, lb, ub)
 
-function (svmsurr::SVMSurrogate)(val::Number)
-    return LIBSVM.predict(svmsurr.model, [val])
-end
+Builds a SVM Surrogate using [LIBSVM](https://github.com/JuliaML/LIBSVM.jl).
 
-"""
-SVMSurrogate(x,y,lb,ub)
+## Arguments
 
-Builds SVM surrogate.
+  - `x`: Input data points.
+  - `y`: Output data points.
+  - `lb`: Lower bound of input data points.
+  - `ub`: Upper bound of output data points.
 """
 function SVMSurrogate(x, y, lb, ub)
-    X = Array{Float64, 2}(undef, length(x), length(x[1]))
-    for j in 1:length(x)
-        X[j, :] = vec(collect(x[j]))
+    X = Array{Float64, 2}(undef, length(x), length(first(x)))
+    if length(lb) == 1
+        for j in eachindex(x)
+            X[j, 1] = x[j]
+        end
+    else
+        for j in eachindex(x)
+            X[j, :] = x[j]
+        end
     end
     model = LIBSVM.fit!(SVC(), X, y)
     SVMSurrogate(x, y, model, lb, ub)
 end
 
 function (svmsurr::SVMSurrogate)(val)
     n = length(val)
-    return LIBSVM.predict(svmsurr.model, reshape(collect(val), 1, n))[1]
+    return LIBSVM.predict(svmsurr.model, reshape(val, 1, n))[1]
 end
 
-function add_point!(svmsurr::SVMSurrogate, x_new, y_new)
+"""
+    update!(svmsurr::SVMSurrogate, x_new, y_new)
+
+## Arguments
+
+  - `svmsurr`: Surrogate of type [`SVMSurrogate`](@ref).
+  - `x_new`: Vector of new data points to be added to the training set of SVMSurrogate.
+  - `y_new`: Vector of new output points to be added to the training set of SVMSurrogate.
+"""
+function update!(svmsurr::SVMSurrogate, x_new, y_new)
+    svmsurr.x = vcat(svmsurr.x, x_new)
+    svmsurr.y = vcat(svmsurr.y, y_new)
     if length(svmsurr.lb) == 1
-        #1D
-        svmsurr.x = vcat(svmsurr.x, x_new)
-        svmsurr.y = vcat(svmsurr.y, y_new)
-        svmsurr.model = LIBSVM.fit!(SVC(), reshape(svmsurr.x, length(svmsurr.x), 1),
-            svmsurr.y)
+        svmsurr.model = LIBSVM.fit!(
+            SVC(), reshape(svmsurr.x, length(svmsurr.x), 1), svmsurr.y)
     else
-        n_previous = length(svmsurr.x)
-        a = vcat(svmsurr.x, x_new)
-        n_after = length(a)
-        dim_new = n_after - n_previous
-        n = length(svmsurr.x)
-        d = length(svmsurr.x[1])
-        tot_dim = n + dim_new
-        X = Array{Float64, 2}(undef, tot_dim, d)
-        for j in 1:n
-            X[j, :] = vec(collect(svmsurr.x[j]))
-        end
-        if dim_new == 1
-            X[n + 1, :] = vec(collect(x_new))
-        else
-            i = 1
-            for j in (n + 1):tot_dim
-                X[j, :] = vec(collect(x_new[i]))
-                i = i + 1
-            end
-        end
-        svmsurr.x = vcat(svmsurr.x, x_new)
-        svmsurr.y = vcat(svmsurr.y, y_new)
-        svmsurr.model = LIBSVM.fit!(SVC(), X, svmsurr.y)
+        svmsurr.model = LIBSVM.fit!(SVC(), transpose(reduce(hcat, svmsurr.x)), svmsurr.y)
     end
-    nothing
 end
+
 end # module

lib/SurrogatesSVM/test/runtests.jl

@@ -1,29 +1,42 @@
 using SafeTestsets
 
 @safetestset "SVMSurrogate" begin
-    using Surrogates, LIBSVM
     using SurrogatesSVM
-
-    #1D
-
-    obj_1D = x -> 2 * x + 1
-    a = 0.0
-    b = 10.0
-    x = sample(5, a, b, SobolSample())
-    y = obj_1D.(x)
-    my_svm_1D = SVMSurrogate(x, y, a, b)
-    val = my_svm_1D(5.0)
-    add_point!(my_svm_1D, 3.1, 7.2)
-    add_point!(my_svm_1D, [3.2, 3.5], [7.4, 8.0])
-
-    #ND
-    obj_N = x -> x[1]^2 * x[2]
-    lb = [0.0, 0.0]
-    ub = [10.0, 10.0]
-    x = sample(100, lb, ub, RandomSample())
-    y = obj_N.(x)
-    my_svm_ND = SVMSurrogate(x, y, lb, ub)
-    val = my_svm_ND((5.0, 1.2))
-    add_point!(my_svm_ND, (1.0, 1.0), 1.0)
-    add_point!(my_svm_ND, [(1.2, 1.2), (1.5, 1.5)], [1.728, 3.375])
+    using Surrogates
+    using LIBSVM
+    using Test
+    @testset "1D" begin
+        obj_1D = x -> 2 * x + 1
+        a = 0.0
+        b = 10.0
+        x = sample(5, a, b, SobolSample())
+        y = obj_1D.(x)
+        svm = LIBSVM.fit!(SVC(), reshape(x, length(x), 1), y)
+        my_svm_1D = SVMSurrogate(x, y, a, b)
+        val = my_svm_1D([5.0])
+        @test LIBSVM.predict(svm, [5.0;;])[1] == val
+        update!(my_svm_1D, [3.1], [7.2])
+        update!(my_svm_1D, [3.2, 3.5], [7.4, 8.0])
+        svm = LIBSVM.fit!(SVC(), reshape(my_svm_1D.x, length(my_svm_1D.x), 1), my_svm_1D.y)
+        val = my_svm_1D([3.1])
+        @test LIBSVM.predict(svm, [3.1;;])[1] == val
+    end
+    @testset "ND" begin
+        obj_N = x -> x[1]^2 * x[2]
+        lb = [0.0, 0.0]
+        ub = [10.0, 10.0]
+        x = collect.(sample(100, lb, ub, RandomSample()))
+        y = obj_N.(x)
+        svm = LIBSVM.fit!(SVC(), transpose(reduce(hcat, x)), y)
+        my_svm_ND = SVMSurrogate(x, y, lb, ub)
+        x_test = [5.0, 1.2]
+        val = my_svm_ND(x_test)
+        @test LIBSVM.predict(svm, reshape(x_test, 1, 2))[1] == val
+        update!(my_svm_ND, [[1.0, 1.0]], [1.0])
+        update!(my_svm_ND, [[1.2, 1.2], [1.5, 1.5]], [1.728, 3.375])
+        svm = LIBSVM.fit!(SVC(), transpose(reduce(hcat, my_svm_ND.x)), my_svm_ND.y)
+        x_test = [1.0, 1.0]
+        val = my_svm_ND(x_test)
+        @test LIBSVM.predict(svm, reshape(x_test, 1, 2))[1] == val
+    end
 end