use rand32 etc

docs/src/tutorials/linear_regression.md

@@ -272,6 +272,8 @@ Let's start by initializing our dataset. We will be using the [`BostonHousing`](
 julia> dataset = BostonHousing();
 
 julia> x, y = BostonHousing(as_df=false)[:];
+
+julia> x, y = Float32.(x), Float32.(y)
 ```
 
 We can now split the obtained data into training and testing data -
@@ -287,7 +289,7 @@ This data contains a diverse number of features, which means that the features h
 
 ```jldoctest linear_regression_complex; filter = r"[+-]?([0-9]*[.])?[0-9]+(f[+-]*[0-9])?"
 julia> std(x_train)
-134.06784844377117
+134.06786f0
 ```
 
 The data is indeed not normalised. We can use the [`Flux.normalise`](@ref) function to normalise the training data.
@@ -296,7 +298,7 @@ The data is indeed not normalised. We can use the [`Flux.normalise`](@ref) funct
 julia> x_train_n = Flux.normalise(x_train);
 
 julia> std(x_train_n)
-1.0000843694328236
+1.0000844f0
 ```
 
 The standard deviation is now close to one! Our data is ready!
@@ -318,7 +320,7 @@ julia> function loss(model, x, y)
        end;
 
 julia> loss(model, x_train_n, y_train)
-676.165591625047
+676.1656f0
 ```
 
 We can now proceed to the training phase!
@@ -363,7 +365,7 @@ Let's have a look at the loss -
 
 ```jldoctest linear_regression_complex; filter = r"[+-]?([0-9]*[.])?[0-9]+(f[+-]*[0-9])?"
 julia> loss(model, x_train_n, y_train)
-27.127200028562164
+27.1272f0
 ```
 
 The loss went down significantly! It can be minimized further by choosing an even smaller `δ`.
@@ -376,7 +378,7 @@ The last step of this tutorial would be to test our model using the testing data
 julia> x_test_n = Flux.normalise(x_test);
 
 julia> loss(model, x_test_n, y_test)
-66.91014769713368
+66.91015f0
 ```
 
 The loss is not as small as the loss of the training data, but it looks good! This also shows that our model is not overfitting!

src/layers/basic.jl

@@ -16,7 +16,7 @@ true
 
 julia> m = Chain(Dense(10 => 5, tanh), Dense(5 => 2));
 
-julia> x = rand(10, 32);
+julia> x = rand32(10, 32);
 
 julia> m(x) == m[2](m[1](x))
 true
@@ -132,11 +132,11 @@ The weight matrix and/or the bias vector (of length `out`) may also be provided
 julia> d = Dense(5 => 2)
 Dense(5 => 2)       # 12 parameters
 
-julia> d(rand(Float32, 5, 64)) |> size
+julia> d(rand32(5, 64)) |> size
 (2, 64)
 
-julia> d(rand(Float32, 5, 1, 1, 64)) |> size  # treated as three batch dimensions
-(2, 1, 1, 64)
+julia> d(rand32(5, 6, 4, 64)) |> size  # treated as three batch dimensions
+(2, 6, 4, 64)
 
 julia> d1 = Dense(ones(2, 5), false, tanh)  # using provided weight matrix
 Dense(5 => 2, tanh; bias=false)  # 10 parameters
@@ -476,7 +476,7 @@ julia> model = Chain(Dense(3 => 5),
                      Parallel(vcat, Dense(5 => 4), Chain(Dense(5 => 7), Dense(7 => 4))),
                      Dense(8 => 17));
 
-julia> model(rand(3)) |> size
+julia> model(rand32(3)) |> size
 (17,)
 
 julia> model2 = Parallel(+; α = Dense(10, 2, tanh), β = Dense(5, 2))
@@ -486,10 +486,10 @@ Parallel(
   β = Dense(5 => 2),                    # 12 parameters
 )                   # Total: 4 arrays, 34 parameters, 392 bytes.
 
-julia> model2(rand(10), rand(5)) |> size
+julia> model2(rand32(10), rand32(5)) |> size
 (2,)
 
-julia> model2[:α](rand(10)) |> size
+julia> model2[:α](rand32(10)) |> size
 (2,)
 
 julia> model2[:β] == model2[2]

src/layers/conv.jl

@@ -22,7 +22,7 @@ See also [`Conv`](@ref), [`MaxPool`](@ref).
 
 # Examples
 ```jldoctest
-julia> xs = rand(Float32, 100, 100, 3, 50);  # a batch of images
+julia> xs = rand32(100, 100, 3, 50);  # a batch of images
 
 julia> layer = Conv((2,2), 3 => 7, pad=SamePad())
 Conv((2, 2), 3 => 7, pad=(1, 0, 1, 0))  # 91 parameters
@@ -96,7 +96,7 @@ See also [`ConvTranspose`](@ref), [`DepthwiseConv`](@ref), [`CrossCor`](@ref).
 
 # Examples
 ```jldoctest
-julia> xs = rand(Float32, 100, 100, 3, 50); # a batch of images
+julia> xs = rand32(100, 100, 3, 50); # a batch of 50 RGB images
 
 julia> layer = Conv((5,5), 3 => 7, relu; bias = false)
 Conv((5, 5), 3 => 7, relu, bias=false)  # 525 parameters
@@ -238,7 +238,7 @@ See also [`Conv`](@ref) for more detailed description of keywords.
 
 # Examples
 ```jldoctest
-julia> xs = rand(Float32, 100, 100, 3, 50);  # a batch of 50 RGB images
+julia> xs = rand32(100, 100, 3, 50);  # a batch of 50 RGB images
 
 julia> layer = ConvTranspose((5,5), 3 => 7, relu)
 ConvTranspose((5, 5), 3 => 7, relu)  # 532 parameters

src/layers/normalise.jl

@@ -131,7 +131,7 @@ Does nothing to the input once [`testmode!`](@ref) is true.
 ```jldoctest
 julia> using Statistics
 
-julia> x = randn(1000,1);
+julia> x = randn32(1000,1);
 
 julia> m = Chain(Dense(1000 => 1000, selu), AlphaDropout(0.2));
 

src/train.jl

@@ -27,10 +27,10 @@ It differs from `Optimisers.setup` in that it:
 
 # Example
 ```jldoctest
-julia> model = Dense(2=>1, leakyrelu; init=ones32);
+julia> model = Dense(2=>1, leakyrelu; init=ones);
 
 julia> opt_state = Flux.setup(Momentum(0.1), model)  # this encodes the optimiser and its state
-(weight = Leaf(Momentum{Float64}(0.1, 0.9), Float32[0.0 0.0]), bias = Leaf(Momentum{Float64}(0.1, 0.9), Float32[0.0]), σ = ())
+(weight = Leaf(Momentum{Float64}(0.1, 0.9), [0.0 0.0]), bias = Leaf(Momentum{Float64}(0.1, 0.9), [0.0]), σ = ())
 
 julia> x1, y1 = [0.2, -0.3], [0.4];  # use the same data for two steps:
 
@@ -39,11 +39,11 @@ julia> Flux.train!(model, [(x1, y1), (x1, y1)], opt_state) do m, x, y
        end
 
 julia> model.bias  # was zero, mutated by Flux.train!
-1-element Vector{Float32}:
- 10.190001
+1-element Vector{Float64}:
+ 10.19
 
 julia> opt_state  # mutated by Flux.train!
-(weight = Leaf(Momentum{Float64}(0.1, 0.9), Float32[-2.018 3.027]), bias = Leaf(Momentum{Float64}(0.1, 0.9), Float32[-10.09]), σ = ())
+(weight = Leaf(Momentum{Float64}(0.1, 0.9), [-2.018 3.027]), bias = Leaf(Momentum{Float64}(0.1, 0.9), [-10.09]), σ = ())
 ```
 """
 function setup(rule::Optimisers.AbstractRule, model)