Added tests for ar arx and fixed bug in ar arx LS_reg

src/SystemIdentification.jl

@@ -32,12 +32,10 @@ typealias PolynomMatrix{T} Union(Array{Polynom{T},1},Polynom{T}, T)
 """
 `a::Vector{Float64}`: The polynomial coeffs A(z) (not including the first 1)\n
 `na::Int`\n
-`λ::Float64`\n
 """
 type AR <: LinearModel
     a::Polynom
     na::Int
-    λ::Float64
 end
 
 """
@@ -45,14 +43,12 @@ end
 `b::VecOrMat{Float64}`: The polynomial coeffs B(z)\n
 `na::Int`\n
 `nb::Vector{Int}`\n
-`λ::Float64`\n
 """
 type ARX <: LinearModel
     a::Polynom
     b::PolynomMatrix
     na::Int
     nb::Polynom{Int}
-    λ::Float64
 end
 
 type RBFARX <: Network
@@ -74,9 +70,10 @@ type FitResult
     error::VecOrMat{Float64}
     fit::FitStatistics
     method::Symbol
-    function FitResult(y,yh,d::Int,method::Symbol)
+    λ::Float64
+    function FitResult(y,yh,d::Int,method::Symbol, λ = 0)
         error = y-yh
-        new(error, FitStatistics(y,yh, d),method)
+        new(error, FitStatistics(y,yh, d),method, λ)
     end
 end
 

src/armax.jl

@@ -5,12 +5,12 @@ function ar(y::Vector{Float64}, na; λ = 0)
     if λ == 0
         w = A\y_train
     else
-        w = (A'A + λeye(size(A,2)))\A'y_train
+        w = (A'A + λ*eye(size(A,2)))\A'y_train
     end
     prediction = A*w
     error = y_train - prediction
-    model = AR(w,na,λ)
-    result = FitResult(y_train,prediction,na, λ>0?(:LS_reg) :(:LS))
+    model = AR(w,na)
+    result = FitResult(y_train,prediction,na, λ>0?(:LS_reg) :(:LS),λ)
     return model, result
 end
 ar(iddata::IdData, na; λ = 0, doplot=false) = ar(iddata.y, na; λ = 0)
@@ -22,7 +22,7 @@ function arx(y::Vector{Float64}, u::VecOrMat{Float64}, na, nb; λ = 0, doplot=fa
     if λ == 0
         w = A\y_train
     else
-        w = (A'A + λeye(size(A,2)))\A'y_train
+        w = (A'A + λ*eye(size(A,2)))\A'y_train
     end
     prediction = A*w
     error = y_train - prediction
@@ -35,8 +35,8 @@ function arx(y::Vector{Float64}, u::VecOrMat{Float64}, na, nb; λ = 0, doplot=fa
         push!(b,w[si:si+nb[i]-1])
         si += nb[i]
     end
-    model = ARX(w[1:na],b,na,nb,λ)
-    result = FitResult(y_train,prediction,na, λ>0?(:LS_reg) :(:LS))
+    model = ARX(w[1:na],b,na,nb)
+    result = FitResult(y_train,prediction,na, λ>0?(:LS_reg) :(:LS), λ)
     return model, result
 end
 arx(iddata::IdData, na, nb; λ = 0) = arx(iddata.y, iddata.u, na, nb; λ = 0)

test/tests.jl

@@ -60,6 +60,12 @@ function run_tests()
         @tassert fit.FIT ≈ 100.0
         @tassert isapprox(fit.RMS, 0.0, atol=1e-10)
         @tassert result.method == :LS
+        @tassert model.na == 2
+        @tassert result.λ == 0
+        model, result = ar(collect(1:5.0),2,λ = 1)
+        @tassert result.method == :LS_reg
+        @tassert result.λ == 1
+
     end
 
     @test "ARX " begin
@@ -68,10 +74,15 @@ function run_tests()
         @tassert isa(model,ARX)
         @tassert isa(result,FitResult)
         @tassert isa(fit,FitStatistics)
-        @tassert model.a ≈ [2;-1]
         @tassert fit.FIT ≈ 100.0
         @tassert isapprox(fit.RMS, 0.0, atol=1e-10)
         @tassert result.method == :LS
+        @tassert model.nb == 1
+        @tassert model.na == 1
+        @tassert result.λ == 0
+        model, result = arx(collect(1:5.0),collect(1:5.0),1,1,λ = 1)
+        @tassert result.method == :LS_reg
+        @tassert result.λ == 1
     end
 end