collect exports

src/Flux.jl

@@ -10,6 +10,11 @@ using DataFlow: graphm, syntax, prewalk!, postwalk!, prewalk, postwalk,
 using DataFlow.Interpreter
 using Juno: Tree, Row
 
+export @net, unroll, unroll1, @shapes,
+  Chain, @Chain, Input, Affine, Conv2D, Recurrent, GatedRecurrent, LSTM,
+  σ, relu, softmax,
+  tf, mxnet
+
 # Zero Flux Given
 
 include("utils.jl")

src/backend/backend.jl

@@ -1,5 +1,3 @@
-export tf, mxnet
-
 # We use a lazy-loading trick to load the backend code as needed; this avoids
 # the need for a hard dependency on both backends.
 

src/compiler/code.jl

@@ -1,8 +1,6 @@
 import DataFlow: cse
 using MacroTools: @q
 
-export @net
-
 function graphdef(ex, params = [])
   @capture(shortdef(ex), (args__,) -> body_)
   body = @> body MacroTools.flatten liftloops graphm DataFlow.il

src/compiler/loops.jl

@@ -1,5 +1,3 @@
-export unroll, unroll1
-
 # Stateful Models
 
 mutable struct Stateful

src/compiler/shape.jl

@@ -1,5 +1,3 @@
-export @shapes
-
 Dims{N} = NTuple{N,Int}
 
 struct Hint
@@ -65,8 +63,6 @@ end
 
 # Shim for kicking off shape inference
 
-export Input
-
 struct Input{N}
   dims::Dims{N}
 end

src/layers/activation.jl

@@ -1,5 +1,3 @@
-export σ, relu, softmax
-
 # Sigmoid
 σ(x) = 1 ./ (1 + exp.(-x))
 back!(::typeof(σ), Δ, x) = Δ .* σ(x).*(1.-σ(x))

src/layers/affine.jl

@@ -1,5 +1,3 @@
-export Affine
-
 @net type Affine
   W
   b

src/layers/control.jl

@@ -1,5 +1,3 @@
-export Chain, @Chain
-
 type Chain
   layers::Vector{Any}
   Chain(xs...) = new([xs...])

src/layers/recurrent.jl

@@ -1,5 +1,3 @@
-export Recurrent, GatedRecurrent, LSTM
-
 @net type Recurrent
   Wxy; Wyy; by
   y

src/layers/shims.jl

@@ -1,5 +1,3 @@
-export Conv2D
-
 struct Conv2D
   filter::Param{Array{Float64,4}} # [height, width, inchans, outchans]
   stride::Dims{2}