Improve array indexing (#142)

src/functions/indAffineIterative.jl

@@ -33,11 +33,10 @@ end
 function prox!(y, f::IndAffineIterative{M, V}, x, gamma) where {M, V}
     # Von Neumann's alternating projections
     R = real(eltype(x))
-    m = size(f.A, 1)
     y .= x
     for k = 1:1000
         maxres = R(0)
-        for i = 1:m
+        for i in eachindex(f.b)
             resi = (f.b[i] - dot(f.A[i,:], y))
             y .= y + resi*f.A[i,:] # no need to divide: rows of A are normalized
             absresi = resi > 0 ? resi : -resi

src/functions/indBallL0.jl

@@ -34,30 +34,28 @@ function (f::IndBallL0)(x)
     return R(0)
 end
 
+function _get_top_k_abs_indices(x::AbstractVector, k)
+    range = firstindex(x):(firstindex(x) + k - 1)
+    return partialsortperm(x, range, by=abs, rev=true)
+end
+
+_get_top_k_abs_indices(x, k) = _get_top_k_abs_indices(x[:], k)
+
 function prox!(y, f::IndBallL0, x, gamma)
     T = eltype(x)
-    p = []
-    if ndims(x) == 1
-        p = partialsortperm(x, 1:f.r, by=abs, rev=true)
-    else
-        p = partialsortperm(x[:], 1:f.r, by=abs, rev=true)
-    end
-    sort!(p)
-    idx = 1
-    for i = 1:length(p)
-        y[idx:p[i]-1] .= T(0)
+    p = _get_top_k_abs_indices(x, f.r)
+    y .= T(0)
+    for i in eachindex(p)
         y[p[i]] = x[p[i]]
-        idx = p[i]+1
     end
-    y[idx:end] .= T(0)
     return real(T)(0)
 end
 
 function prox_naive(f::IndBallL0, x, gamma)
     T = eltype(x)
     p = sortperm(abs.(x)[:], rev=true)
     y = similar(x)
-    y[p[1:f.r]] .= x[p[1:f.r]]
-    y[p[f.r+1:end]] .= T(0)
+    y[p[begin:begin+f.r-1]] .= x[p[begin:begin+f.r-1]]
+    y[p[begin+f.r:end]] .= T(0)
     return y, real(T)(0)
 end

src/functions/indPSD.jl

@@ -94,7 +94,7 @@ function (f::IndPSD)(x::AbstractVector{Float64})
     f.scaling && scale_diagonal!(y, sqrt(2))
 
     Z = dspev!(:N, :L, y)
-    for i in 1:length(Z)
+    for i in eachindex(Z)
         # Do we allow for some tolerance here?
         if Z[i] <= -1e-14
             return +Inf
@@ -118,7 +118,7 @@ function prox!(y::AbstractVector{Float64}, f::IndPSD, x::AbstractVector{Float64}
     # Now let M = Z*diagm(W)*Z'
     M = M*Z'
     n = length(W)
-    k = 1
+    k = firstindex(y)
     # Store lower diagonal of M in y
     for j in 1:n, i in j:n
         y[k] = M[i,j]
@@ -135,8 +135,8 @@ function prox_naive(f::IndPSD, x::AbstractVector{Float64}, gamma)
     # Formula for size of matrix
     n = Int(sqrt(1/4+2*length(x))-1/2)
     X = Matrix{Float64}(undef, n, n)
-    k = 1
-    # Store y in M
+    k = firstindex(x)
+    # Store x in X
     for j = 1:n, i = j:n
         # Lower half
         X[i,j] = x[k]
@@ -164,7 +164,7 @@ function prox_naive(f::IndPSD, x::AbstractVector{Float64}, gamma)
     end
 
     y = similar(x)
-    k = 1
+    k = firstindex(y)
     # Store Lower half of X in y
     for j = 1:n, i = j:n
         y[k] = X[i,j]

src/functions/normL21.jl

@@ -33,17 +33,17 @@ function (f::NormL21)(X)
     nslice = R(0)
     n21X = R(0)
     if f.dim == 1
-        for j = 1:size(X, 2)
+        for j in axes(X, 2)
             nslice = R(0)
-            for i = 1:size(X, 1)
+            for i in axes(X, 1)
                 nslice += abs(X[i, j])^2
             end
             n21X += sqrt(nslice)
         end
     elseif f.dim == 2
-        for i = 1:size(X, 1)
+        for i in axes(X, 1)
             nslice = R(0)
-            for j = 1:size(X, 2)
+            for j in axes(X, 2)
                 nslice += abs(X[i, j])^2
             end
             n21X += sqrt(nslice)
@@ -58,29 +58,29 @@ function prox!(Y, f::NormL21, X, gamma)
     nslice = R(0)
     n21X = R(0)
     if f.dim == 1
-        for j = 1:size(X, 2)
+        for j in axes(X, 2)
             nslice = R(0)
-            for i = 1:size(X, 1)
+            for i in axes(X, 1)
                 nslice += abs(X[i, j])^2
             end
             nslice = sqrt(nslice)
             scal = 1 - gl / nslice
             scal = scal <= 0 ? R(0) : scal
-            for i = 1:size(X, 1)
+            for i in axes(X, 1)
                 Y[i, j] = scal * X[i, j]
             end
             n21X += scal * nslice
         end
     elseif f.dim == 2
-        for i = 1:size(X, 1)
+        for i in axes(X, 1)
             nslice = R(0)
-            for j = 1:size(X, 2)
+            for j in axes(X, 2)
                 nslice += abs(X[i, j])^2
             end
             nslice = sqrt(nslice)
             scal = 1-gl/nslice
             scal = scal <= 0 ? R(0) : scal
-            for j = 1:size(X, 2)
+            for j in axes(X, 2)
                 Y[i, j] = scal * X[i, j]
             end
             n21X += scal * nslice

test/test_calculus.jl

@@ -100,11 +100,11 @@ stuff = [
       )
 ]
 
-@testset "$i" for i = 1:length(stuff)
+@testset "$i" for i in eachindex(stuff)
     f = stuff[i]["funcs"][1]
     g = stuff[i]["funcs"][2]
 
-    for j = 1:length(stuff[i]["args"])
+    for j in eachindex(stuff[i]["args"])
         x = stuff[i]["args"][j]
         gamma = stuff[i]["gammas"][j]
 

test/test_gradients.jl

@@ -141,7 +141,7 @@ stuff = [
       ),
 ]
 
-for i = 1:length(stuff)
+for i in eachindex(stuff)
 
   f = stuff[i]["f"]
   x = stuff[i]["x"]

test/test_graph.jl

@@ -66,17 +66,17 @@ stuff = [
       ),
 ]
 
-for i = 1:length(stuff)
+for i in eachindex(stuff)
   constr = stuff[i]["constr"]
 
   if haskey(stuff[i], "wrong")
-    for j = 1:length(stuff[i]["wrong"])
+    for j in eachindex(stuff[i]["wrong"])
       wrong = stuff[i]["wrong"][j]
       @test_throws ErrorException constr(wrong...)
     end
   end
 
-  for j = 1:length(stuff[i]["params"])
+  for j in eachindex(stuff[i]["params"])
     params = stuff[i]["params"][j]
     x      = stuff[i]["args"][j]
     f = constr(params...)

test/test_results.jl

@@ -321,7 +321,7 @@ stuff = [
       )
 ]
 
-@testset "$(i)" for i = 1:length(stuff)
+@testset "$(i)" for i in eachindex(stuff)
 
     f = stuff[i]["f"]
     x = stuff[i]["x"]