spacing

Author: cossio

src/utils.jl

@@ -285,7 +285,7 @@ sparse_init(rng::AbstractRNG; init_kwargs...) = (dims...; kwargs...) -> sparse_i
 """
     identity_init([rng=GLOBAL_RNG], dims...; gain=1, shift=0)
 
-Return an `Array` of size `dims` which yields an identity mapping when used as parameters in 
+Return an `Array` of size `dims` which yields an identity mapping when used as parameters in
 most Flux layers. Use `gain` to scale the identity by a constant.
 
 Often useful in the context of transfer learning, i.e when one wants to add more capacity to
@@ -297,10 +297,10 @@ Equivalent to `Base.circshift(identity(dims...), shift)`.
 Some caveats: Not all layers will be identity mapping when used with this init. Exceptions
 include recurrent layers, `DepthwiseConv` and normalization layers.
 
-Also note that layers must have `input_size == output_size` for identity mapping to be 
+Also note that layers must have `input_size == output_size` for identity mapping to be
 possible. When this is not the case, extra dimensions of the array are padded with zeros.
 
-For convolutional layers, in addition to the above, the kernel sizes must also be odd and 
+For convolutional layers, in addition to the above, the kernel sizes must also be odd and
 padding must be applied so that output feature maps have the same size as input feature maps,
 e.g by using [`SamePad`](@ref).
 
@@ -574,7 +574,7 @@ See also [`unstack`](@ref).
 # Examples
 
 ```jldoctest
-julia> Flux.unbatch([1 3 5 7; 
+julia> Flux.unbatch([1 3 5 7;
                      2 4 6 8])
 4-element Vector{Vector{Int64}}:
  [1, 2]

test/utils.jl

@@ -1,6 +1,6 @@
 using Flux
 using Flux: throttle, nfan, glorot_uniform, glorot_normal,
-             kaiming_normal, kaiming_uniform, orthogonal, 
+             kaiming_normal, kaiming_uniform, orthogonal,
              sparse_init, stack, unstack, Zeros, batch, unbatch
 using StatsBase: var, std
 using Random
@@ -178,10 +178,10 @@ end
 
     @testset "$layer ID mapping with kernelsize $kernelsize" for layer in (Conv, ConvTranspose, CrossCor), kernelsize in (
         (1,),
-        (3,), 
-        (1, 3), 
-        (3, 5), 
-        (3, 5, 7))   
+        (3,),
+        (1, 3),
+        (3, 5),
+        (3, 5, 7))
         nch = 3
         l = layer(kernelsize, nch=>nch, init=identity_init, pad=SamePad())
 
@@ -333,9 +333,9 @@ end
 
 
 @testset "Batching" begin
-  stacked_array=[ 8 9 3 5 
-                  9 6 6 9 
-                  9 1 7 2 
+  stacked_array=[ 8 9 3 5
+                  9 6 6 9
+                  9 1 7 2
                   7 4 10 6 ]
   unstacked_array=[[8, 9, 9, 7], [9, 6, 1, 4], [3, 6, 7, 10], [5, 9, 2, 6]]
   @test unbatch(stacked_array) == unstacked_array
@@ -445,7 +445,7 @@ end
 
   modules = Flux.modules(Chain(SkipConnection(
                                   Conv((2,3), 4=>5; pad=6, stride=7),
-                                  +), 
+                                  +),
                                 LayerNorm(8)))
   @test length(modules) == 5
 end
@@ -475,16 +475,16 @@ end
   @testset "early stopping" begin
     @testset "args & kwargs" begin
       es = Flux.early_stopping((x; y = 1) -> x + y, 10; min_dist=3)
-  
+
       n_iter = 0
       while n_iter < 99
         es(-n_iter; y=-n_iter) && break
         n_iter += 1
       end
-  
+
       @test n_iter == 9
     end
-  
+
     @testset "distance" begin
       es = Flux.early_stopping(identity, 10; distance=(best_score, score) -> score - best_score)
 
@@ -496,7 +496,7 @@ end
 
       @test n_iter == 99
     end
-  
+
     @testset "init_score" begin
       es = Flux.early_stopping(identity, 10; init_score=10)