using axes across codebase

src/aggregations/segmented_lse.jl

@@ -125,7 +125,7 @@ end
 
 function segmented_lse_back(Δ, ::Missing, ψ, bags)
     dψ = zero(ψ)
-    @inbounds for (bi, b) in enumerate(bags)
+    @inbounds for bi in eachindex(bags)
         for i in eachindex(ψ)
             dψ[i] += Δ[i, bi]
         end

src/aggregations/segmented_max.jl

@@ -46,7 +46,7 @@ function segmented_max_forw(x::AbstractMatrix, ψ::AbstractVector, bags::Abstrac
             end
         else
             for j in b
-                for i in 1:size(x, 1)
+                for i in axes(x, 1)
                     y[i, bi] = max(y[i, bi], x[i, j])
                 end
             end
@@ -67,20 +67,17 @@ function segmented_max_back(Δ, y, x, ψ, bags)
             end
         else
             fi = first(b)
-            v .= @view x[:,fi]
-            # for k in axes(v,1)
-            #     v[k] = x[k,fi]
-            # end
+            v .= @view x[:, fi]
             idxs .= fi
             for j in b
-                for i in axes(x,1)
+                for i in axes(x, 1)
                     if v[i] < x[i, j]
                         idxs[i] = j
                         v[i] = x[i, j]
                     end
                 end
             end
-            for i in axes(x,1)
+            for i in axes(x, 1)
                 dx[i, idxs[i]] += Δ[i, bi]
             end
         end
@@ -90,7 +87,7 @@ end
 
 function segmented_max_back(Δ, y, x::Missing, ψ, bags) 
     dψ = zero(ψ)
-    @inbounds for (bi, b) in enumerate(bags)
+    @inbounds for bi in eachindex(bags)
         for i in eachindex(ψ)
             dψ[i] += Δ[i, bi]
         end

src/aggregations/segmented_mean.jl

@@ -46,7 +46,7 @@ function segmented_mean_forw(x::AbstractMatrix, ψ::AbstractVector, bags::Abstra
             end
         else
             for j in b
-                for i in 1:size(x, 1)
+                for i in axes(x, 1)
                     y[i, bi] += _weight(w, i, j, t) * x[i, j]
                 end
             end
@@ -68,7 +68,7 @@ function segmented_mean_back(Δ, y, x, ψ, bags, w)
         else
             ws = _bagnorm(w, b)
             for j in b
-                for i in 1:size(x, 1)
+                for i in axes(x, 1)
                     dx[i, j] += _weight(w, i, j, eltype(x)) * Δ[i, bi] / _weightsum(ws, i)
                     ∇dw_segmented_mean!(dw, Δ, x, y, w, ws, i, j, bi)
                 end
@@ -80,7 +80,7 @@ end
 
 function segmented_mean_back(Δ, y, x::Missing, ψ, bags, w) 
     dψ = zero(ψ)
-    @inbounds for (bi, b) in enumerate(bags)
+    @inbounds for bi in eachindex(bags)
         for i in eachindex(ψ)
             dψ[i] += Δ[i, bi]
         end

src/aggregations/segmented_pnorm.jl

@@ -55,7 +55,7 @@ function _pnorm_precomp(x::AbstractMatrix, bags)
     @inbounds for (bi, b) in enumerate(bags)
         if !isempty(b)
             for j in b
-                for i in 1:size(x, 1)
+                for i in axes(x, 1)
                     M[i, bi] = max(M[i, bi], abs(x[i, j]))
                 end
             end
@@ -77,11 +77,11 @@ function _segmented_pnorm_norm(a::AbstractMatrix, ψ::AbstractVector, p::Abstrac
         else
             ws = _bagnorm(w, b)
             for j in b
-                for i in 1:size(a, 1)
+                for i in axes(a, 1)
                     y[i, bi] += _weight(w, i, j, t) * abs(a[i, j] / M[i, bi]) ^ p[i]
                 end
             end
-            for i in 1:size(a, 1)
+            for i in axes(a, 1)
                 y[i, bi] = M[i, bi] * (y[i, bi] / _weightsum(ws, i)) ^ (one(t) / p[i])
             end
         end
@@ -97,9 +97,7 @@ end
 
 function segmented_pnorm_back(Δ, y, a, ψ, p, bags, w, M)
     da = zero(a)
-    dp = zero(p)
-    dps1 = zero(p)
-    dps2 = zero(p)
+    dp, dps1, dps2 = zero(p), zero(p), zero(p)
     dψ = zero(ψ)
     @inbounds for (bi, b) in enumerate(bags)
         if isempty(b)
@@ -111,7 +109,7 @@ function segmented_pnorm_back(Δ, y, a, ψ, p, bags, w, M)
             dps1 .= zero(eltype(p))
             dps2 .= zero(eltype(p))
             for j in b
-                for i in 1:size(a, 1)
+                for i in axes(a, 1)
                     ab = abs(a[i, j])
                     da[i, j] += Δ[i, bi] * _weight(w, i, j, eltype(p)) *
                         sign(a[i, j]) / _weightsum(ws, i) * (ab / y[i, bi]) ^ (p[i] - one(eltype(p)))
@@ -120,7 +118,7 @@ function segmented_pnorm_back(Δ, y, a, ψ, p, bags, w, M)
                     dps2[i] += ww
                 end
             end
-            for i in 1:size(a, 1)
+            for i in axes(a, 1)
                 t = y[i, bi] / p[i]
                 t *= dps1[i] / dps2[i] - (p[i] * log(M[i, bi]) + log(dps2[i]) - log(_weightsum(ws, i))) / p[i]
                 dp[i] += Δ[i, bi] * t
@@ -132,7 +130,7 @@ end
 
 function segmented_pnorm_back(Δ, y, ψ, bags) 
     dψ = zero(ψ)
-    @inbounds for (bi, b) in enumerate(bags)
+    @inbounds for bi in eachindex(bags)
         for i in eachindex(ψ)
             dψ[i] += Δ[i, bi]
         end

src/aggregations/segmented_sum.jl

@@ -46,7 +46,7 @@ function segmented_sum_forw(x::AbstractMatrix, ψ::AbstractVector, bags::Abstrac
             end
         else
             for j in b
-                for i in 1:size(x, 1)
+                for i in axes(x, 1)
                     y[i, bi] += _weight(w, i, j, t) * x[i, j]
                 end
             end
@@ -66,7 +66,7 @@ function segmented_sum_back(Δ, y, x, ψ, bags, w)
             end
         else
             for j in b
-                for i in 1:size(x, 1)
+                for i in axes(x, 1)
                     dx[i, j] += _weight(w, i, j, eltype(x)) * Δ[i, bi]
                     ∇dw_segmented_sum!(dw, Δ, x, y, w, i, j, bi)
                 end

src/conv.jl

@@ -12,8 +12,8 @@ _subsetof(bags) = bags[_getrange(length(bags))]
 _convshift(n) = (i = div(n, 2); mod(n, 2) == 0 ? (1 - i:i) : (-i : i) )
 
 function _addmatvec!(o::Matrix, i, W::Matrix, x::Matrix, j)
-    @inbounds for s in 1:size(W, 2)
-        for r in 1:size(W, 1)
+    @inbounds for s in axes(W, 2)
+        for r in axes(W, 1)
             o[r, i] += W[r, s] * x[s, j]
         end
     end
@@ -25,23 +25,23 @@ function _addmatvec!(o::Matrix, i, W::Matrix, x::SparseMatrixCSC, j)
     rowptr = x.rowval[rn]
     vals = x.nzval[rn]
     @inbounds for (s, v) in zip(rowptr, vals)
-        for r in 1:size(W, 1)
+        for r in axes(W, 1)
             o[r, i] += W[r, s] * v
         end
     end
 end
 
 function _addmattvec!(o::Matrix, i, W::Matrix, x::AbstractMatrix, j)
-    @inbounds for s in 1:size(W, 1)
-        for r in 1:size(W, 2)
+    @inbounds for s in axes(W, 1)
+        for r in axes(W, 2)
             o[r, i] += W[s, r] * x[s, j]
         end
     end
 end
 
 function _addvecvect!(W::Matrix, Δ::Matrix, i, x::AbstractMatrix, j)
-    @inbounds for r in 1:size(W, 2)
-        for s in 1:size(W, 1)
+    @inbounds for r in axes(W, 2)
+        for s in axes(W, 1)
             W[s, r] += Δ[s, i] * x[r, j]
         end
     end
@@ -53,7 +53,7 @@ function _addvecvect!(W::Matrix, Δ::Matrix, i, x::SparseMatrixCSC, j)
     rowptr = x.rowval[rn]
     vals = x.nzval[rn]
     @inbounds for (r, v) in zip(rowptr, vals)
-        for s in 1:size(W, 1)
+        for s in axes(W, 1)
             W[s, r] += Δ[s, i] * v
         end
     end
@@ -257,4 +257,4 @@ function ChainRulesCore.rrule(::typeof(convsum), bags, xs...)
     convsum(bags, xs...), Δ -> (NoTangent(), NoTangent(),  ∇convsum(Δ, bags, length(xs))...)
 end
 
-legacy_bagconv(x, bags, f::AbstractArray{T, 3}) where {T} = convsum(bags, [f[:, :, i]*x for i in 1:size(f, 3)]...)
+legacy_bagconv(x, bags, f::AbstractArray{T, 3}) where {T} = convsum(bags, [f[:, :, i]*x for i in axes(f, 3)]...)

src/special_arrays/ngram_matrix.jl

@@ -338,12 +338,12 @@ end
 SparseArrays.SparseMatrixCSC(x::NGramMatrix) = SparseArrays.SparseMatrixCSC{Int64, UInt}(x)
 function SparseArrays.SparseMatrixCSC{Tv, Ti}(x::NGramMatrix) where {Tv, Ti <: Integer}
     size(x, 2) == 0 && return sparse(Ti[],Ti[],Tv[], size(x,1), size(x,2))
-    l = sum(map(i -> length(NGramIterator(x, i)), 1:size(x, 2)))
+    l = sum(map(i -> length(NGramIterator(x, i)), axes(x, 2)))
     I = zeros(Ti, l)
     J = zeros(Ti, l)
     V = ones(Tv, l)
     vid = 1
-    for j in 1:size(x, 2)
+    for j in axes(x, 2)
         for i in NGramIterator(x, j)
             I[vid] = i + 1
             J[vid] = j
@@ -377,7 +377,7 @@ function _mul_ngram_vec!(s, A, B, bn, C, k, z, ψ=nothing)
     for l in 1:_len(s, B.n)
         z = mod(_next_ngram(z, l, s, B.n, B.b, bn), B.m)
         zi = z + 1
-        for i in 1:size(C, 1)
+        for i in axes(C, 1)
             @inbounds C[i, k] += A[i, zi]
         end
     end
@@ -403,7 +403,7 @@ function _∇A_mul_ngram_vec!(Δ, s, B, bn, ∇A, k, z)
     for l in 1:_len(s, B.n)
         z = mod(_next_ngram(z, l, s, B.n, B.b, bn), B.m)
         zi = z + 1
-        for i in 1:size(∇A, 1)
+        for i in axes(∇A, 1)
             @inbounds ∇A[i, zi] += Δ[i, k]
         end
     end

src/special_arrays/postimputing_matrix.jl

@@ -89,7 +89,7 @@ function _postimpute_maybe_hot(A, B)
     @inbounds for (j, k) in enumerate(B.I)
         if !ismissing(k)
             m[j] = false
-            for i in 1:size(C, 1)
+            for i in axes(C, 1)
                 C[i, j] = A.W[i, k]
             end
         end
@@ -103,7 +103,7 @@ function ChainRulesCore.rrule(::typeof(_mul_pi_maybe_hot), A, B)
         dW = zero(A.W)
         @inbounds for (k, j) in enumerate(B.I)
             if !ismissing(j)
-                for i in 1:size(dW, 1)
+                for i in axes(dW, 1)
                     dW[i, j] += Δ[i, k]
                 end
             end