Remove redundant randomness, add docfilters, and make them stricter for Dropout layer

Author: Saransh-cpp

src/layers/normalise.jl

@@ -67,7 +67,7 @@ Custom RNGs are only supported on the CPU.
 Does nothing to the input once [`Flux.testmode!`](@ref) is `true`.
 
 # Examples
-```jldoctest
+```jldoctest; filter = r"[+-]?(?:(?:[0-9])(?:\\.\\d+)?)|(?:1)(?:\\.0+)?"
 julia> m = Chain(Dense(2 => 2), Dropout(1))
 Chain(
   Dense(2 => 2),                        # 6 parameters
@@ -136,14 +136,10 @@ remain the same as before.
 Does nothing to the input once [`testmode!`](@ref) is true.
 
 # Examples
-```jldoctest
+```jldoctest; filter = r"[+-]?([0-9]*[.])?[0-9]+"
 julia> x = randn(20,1);
 
-julia> m = Chain(Dense(20 => 10, selu), AlphaDropout(0.5))
-Chain(
-  Dense(20 => 10, selu),                # 210 parameters
-  AlphaDropout{Float64, Random.TaskLocalRNG}(0.5, nothing, Random.TaskLocalRNG()),
-)
+julia> m = Chain(Dense(20 => 10, selu), AlphaDropout(0.5));
 
 julia> Flux.trainmode!(m);
 
@@ -212,7 +208,7 @@ using the [`Scale`](@ref) layer.
 See also [`BatchNorm`](@ref), [`InstanceNorm`](@ref), [`GroupNorm`](@ref), and [`normalise`](@ref).
 
 # Examples
-```jldoctest
+```jldoctest; filter = r"[+-]?([0-9]*[.])?[0-9]+"
 julia> xs = rand(3, 3, 3, 2);  # a batch of 2 3X3X3 images
 
 julia> m = LayerNorm(3);
@@ -423,7 +419,7 @@ that will be used to renormalize the input in test phase.
 in previous Flux versions (< v0.12).
 
 # Examples
-```jldoctest
+```jldoctest; filter = r"[+-]?([0-9]*[.])?[0-9]+"
 julia> xs = rand(3, 3, 3, 2);  # a batch of 2 3X3X3 images
 
 julia> m = InstanceNorm(3);
@@ -521,7 +517,7 @@ If `track_stats=true`, accumulates mean and var statistics in training phase
 that will be used to renormalize the input in test phase.
 
 # Examples
-```jldoctest
+```jldoctest; filter = r"[+-]?([0-9]*[.])?[0-9]+"
 julia> xs = rand(3, 3, 4, 2);  # a batch of 2 3X3X4 images
 
 julia> m = GroupNorm(4, 2);

src/layers/recurrent.jl

@@ -180,31 +180,23 @@ Assuming you have a `Recur` layer `rnn`, this is roughly equivalent to:
     rnn.state = hidden(rnn.cell)
 
 # Examples
-```jldoctest
-julia> r = RNN(3 => 5);
+```jldoctest; filter = r"[+-]?([0-9]*[.])?[0-9]+"
+julia> r = RNN(1 => 1);
+
+julia> a = Vector{Float32}([1])
+1-element Vector{Float32}:
+ 1.0
 
 julia> r.state
-5×1 Matrix{Float32}:
- 0.0
- 0.0
- 0.0
- 0.0
+1×1 Matrix{Float32}:
  0.0
 
-julia> r(rand(Float32, 3)); r.state
-5×1 Matrix{Float32}:
- -0.32719195
- -0.45280662
- -0.50386846
- -0.14782222
-  0.23584609
+julia> r(a); r.state
+1×1 Matrix{Float32}:
+ 0.61431444
 
 julia> Flux.reset!(r)
-5×1 Matrix{Float32}:
- 0.0
- 0.0
- 0.0
- 0.0
+1×1 Matrix{Float32}:
  0.0
 ```
 """

src/layers/upsample.jl

@@ -81,34 +81,13 @@ resolution images while upscaling them.
 See [`NNlib.pixel_shuffle`](@ref).
 
 # Examples
-```jldoctest; filter = r"[+-]?([0-9]*[.])?[0-9]+"
+```jldoctest
 julia> p = PixelShuffle(2);
 
-julia> xs = rand(2, 2, 4, 1)  # an image with 4 channels having 2X2 pixels in each channel
-2×2×4×1 Array{Float64, 4}:
-[:, :, 1, 1] =
- 0.826452   0.0519244
- 0.0686387  0.438346
-
-[:, :, 2, 1] =
- 0.343179  0.445101
- 0.543927  0.740905
-
-[:, :, 3, 1] =
- 0.105997  0.422996
- 0.32957   0.167205
-
-[:, :, 4, 1] =
- 0.825737  0.98609
- 0.757365  0.294784
-
-julia> p(xs)  # upsampled image with only 1 channel
-4×4×1×1 Array{Float64, 4}:
-[:, :, 1, 1] =
- 0.826452   0.105997  0.0519244  0.422996
- 0.343179   0.825737  0.445101   0.98609
- 0.0686387  0.32957   0.438346   0.167205
- 0.543927   0.757365  0.740905   0.294784
+julia> xs = rand(2, 2, 4, 1); # an image with 4 channels having 2X2 pixels in each channel
+
+julia> p(xs) |> size  # upsampled image with only 1 channel
+(4, 4, 1, 1)
 ```
 """
 struct PixelShuffle