use gather; fix outdated docs

Co-authored-by: Manikya <manikyabard@gmail.com>

docs/src/gpu.md

@@ -30,7 +30,7 @@ If you define a structured model, like a `Dense` layer or `Chain`, you just need
 ```julia
 d = Dense(10, 5, σ)
 d = fmap(cu, d)
-d.W # CuArray
+d.weight # CuArray
 d(cu(rand(10))) # CuArray output
 
 m = Chain(Dense(10, 5, σ), Dense(5, 2), softmax)

docs/src/models/advanced.md

@@ -68,7 +68,7 @@ by simply deleting it from `ps`:
 
 ```julia
 ps = params(m)
-delete!(ps, m[2].b) 
+delete!(ps, m[2].bias) 
 ```
 
 ## Custom multiple input or output layer

docs/src/models/nnlib.md

@@ -67,3 +67,10 @@ NNlib.batched_mul!
 NNlib.batched_adjoint
 NNlib.batched_transpose
 ```
+
+## Gather and Scatter
+
+```@docs
+NNlib.gather
+NNlib.scatter
+```

docs/src/models/overview.md

@@ -15,7 +15,7 @@ Here's how you'd use Flux to build and train the most basic of models, step by s
 
 This example will predict the output of the function `4x + 2`. First, import `Flux` and define the function we want to simulate:
 
-```
+```julia
 julia> using Flux
 
 julia> actual(x) = 4x + 2
@@ -28,7 +28,7 @@ This example will build a model to approximate the `actual` function.
 
 Use the `actual` function to build sets of data for training and verification:
 
-```
+```julia
 julia> x_train, x_test = hcat(0:5...), hcat(6:10...)
 ([0 1 … 4 5], [6 7 … 9 10])
 
@@ -42,38 +42,38 @@ Normally, your training and test data come from real world observations, but thi
 
 Now, build a model to make predictions with `1` input and `1` output:
 
-```
+```julia
 julia> model = Dense(1, 1)
 Dense(1, 1)
 
-julia> model.W
-1-element Array{Float64,1}:
- -0.99009055
+julia> model.weight
+1×1 Matrix{Float32}:
+ -1.4925033
 
-julia> model.b
-1-element Array{Float64,1}:
+julia> model.bias
+1-element Vector{Float32}:
  0.0
 ```
 
-Under the hood, a dense layer is a struct with fields `W` and `b`. `W` represents a weight and `b` represents a bias. There's another way to think about a model. In Flux, *models are conceptually predictive functions*: 
+Under the hood, a dense layer is a struct with fields `weight` and `bias`. `weight` represents a weights' matrix and `bias` represents a bias vector. There's another way to think about a model. In Flux, *models are conceptually predictive functions*: 
 
-```
+```julia
 julia> predict = Dense(1, 1)
 ```
 
 `Dense(1, 1)` also implements the function `σ(Wx+b)` where `W` and `b` are the weights and biases. `σ` is an activation function (more on activations later). Our model has one weight and one bias, but typical models will have many more. Think of weights and biases as knobs and levers Flux can use to tune predictions. Activation functions are transformations that tailor models to your needs. 
 
 This model will already make predictions, though not accurate ones yet:
 
-```
+```julia
 julia> predict(x_train)
-1×6 Array{Float32,2}:
- -1.98018  -5.94054  -9.90091  -13.8613  -17.8216  -21.782
+1×6 Matrix{Float32}:
+ 0.0  -1.4925  -2.98501  -4.47751  -5.97001  -7.46252
 ```
 
 In order to make better predictions, you'll need to provide a *loss function* to tell Flux how to objectively *evaluate* the quality of a prediction. Loss functions compute the cumulative distance between actual values and predictions. 
 
-```
+```julia
 julia> loss(x, y) = Flux.Losses.mse(predict(x), y)
 loss (generic function with 1 method)
 
@@ -87,7 +87,7 @@ More accurate predictions will yield a lower loss. You can write your own loss f
 
 Under the hood, the Flux [`train!`](@ref) function uses *a loss function* and *training data* to improve the *parameters* of your model based on a pluggable [`optimiser`](../training/optimisers.md):
 
-```
+```julia
 julia> using Flux: train!
 
 julia> opt = Descent()
@@ -100,12 +100,12 @@ julia> data = [(x_train, y_train)]
 
 Now, we have the optimiser and data we'll pass to `train!`. All that remains are the parameters of the model. Remember, each model is a Julia struct with a function and configurable parameters. Remember, the dense layer has weights and biases that depend on the dimensions of the inputs and outputs: 
 
-```
-julia> predict.W
+```julia
+julia> predict.weight
 1-element Array{Float64,1}:
  -0.99009055
 
-julia> predict.b
+julia> predict.bias
 1-element Array{Float64,1}:
  0.0
 ```
@@ -120,7 +120,7 @@ Params([[-0.99009055], [0.0]])
 These are the parameters Flux will change, one step at a time, to improve predictions. Each of the parameters comes from the `predict` model: 
 
 ```
-julia> predict.W in parameters, predict.b in parameters
+julia> predict.weight in parameters, predict.bias in parameters
 (true, true)
 
 ```

docs/src/models/regularisation.md

@@ -13,10 +13,10 @@ m = Dense(10, 5)
 loss(x, y) = logitcrossentropy(m(x), y)
 ```
 
-We can apply L2 regularisation by taking the squared norm of the parameters , `m.W` and `m.b`.
+We can apply L2 regularisation by taking the squared norm of the parameters , `m.weight` and `m.bias`.
 
 ```julia
-penalty() = sum(abs2, m.W) + sum(abs2, m.b)
+penalty() = sum(abs2, m.weight) + sum(abs2, m.bias)
 loss(x, y) = logitcrossentropy(m(x), y) + penalty()
 ```
 

src/layers/basic.jl

@@ -475,7 +475,8 @@ function Embedding(in::Integer, out::Integer;
 end
 
 (m::Embedding)(x::Union{OneHotVector, OneHotMatrix}) = m.weight * x # equivalent to m.weight[:,onecold(x)]
-(m::Embedding)(x::Union{Int,AbstractVector}) = m.weight[:, x]
+(m::Embedding)(x::Integer) = m([x])
+(m::Embedding)(x::AbstractVector) = NNlib.gather(m.weight, x)
 (m::Embedding)(x::AbstractArray) = reshape(m(vec(x)), :, size(x)...)
 
 function Base.show(io::IO, m::Embedding)

src/utils.jl

@@ -15,7 +15,7 @@ This function is mainly used by weight initializers, e.g., [`kaiming_normal`](@r
 ```jldoctest
 julia> layer = Dense(10, 20);
 
-julia> Flux.nfan(size(layer.W))
+julia> Flux.nfan(size(layer.weight))
 (10, 20)
 
 julia> layer = Conv((3, 3), 2=>10);

test/cuda/layers.jl

@@ -140,7 +140,7 @@ end
 
   @test sum(l(ip)) ≈ 0.f0
   gs = gradient(() -> sum(l(ip)), Flux.params(l))
-  @test l.b ∉ gs.params
+  @test l.bias ∉ gs.params
 end
 
 @testset "Extended BatchNorm" begin

test/utils.jl

@@ -226,19 +226,19 @@ end
   m = Chain(Dense(10, 5, relu), Dense(5, 2))
   x64 = rand(Float64, 10)
   x32 = rand(Float32, 10)
-  @test eltype(m[1].W) == Float32
+  @test eltype(m[1].weight) == Float32
   @test eltype(m(x32)) == Float32
   @test eltype(m(x64)) == Float64
   @test eltype(f64(m)(x32)) == Float64
   @test eltype(f64(m)(x64)) == Float64
-  @test eltype(f64(m)[1].W) == Float64
-  @test eltype(f32(f64(m))[1].W) == Float32
+  @test eltype(f64(m)[1].weight) == Float64
+  @test eltype(f32(f64(m))[1].weight) == Float32
 end
 
 @testset "Zeros" begin
   m = Dense(3,2; bias=false)
-  @test f64(m).b === m.b === Zeros()
-  @test f32(m).b === m.b === Zeros()
+  @test f64(m).bias === m.bias === Zeros()
+  @test f32(m).bias === m.bias === Zeros()
 
   @testset "Gradients for broadcasted $op with sizes $s" for op in (+,-,*), s in ((1,), (2,3))
     o = ones(s)
@@ -340,19 +340,19 @@ end
 
   nobias(n) = Zeros()
   testdense(m, bt) = @testset "Check layer $i" for (i, (l1, l2)) in enumerate(zip(m, dm(bt)))
-    @test l1.W == l2.W
-    @test l1.b == l2.b
-    @test_skip typeof(l1.b) === typeof(l2.b)
+    @test l1.weight == l2.weight
+    @test l1.bias == l2.bias
+    @test_skip typeof(l1.bias) === typeof(l2.bias)
   end
 
   @testset "loadparams!" begin
     import Flux: loadparams!
     pars(w, b) = [w, b]
     import Flux: loadparams!, Zeros
     pars(w, b::Zeros) = [w, Flux.zeros(size(w,1))]
-    pars(l) = pars(l.W, l.b)
+    pars(l) = pars(l.weight, l.bias)
     pararray(m) = mapreduce(pars, vcat, m)
-    weights(m) = mapreduce(l -> [l.W], vcat, m)
+    weights(m) = mapreduce(l -> [l.weight], vcat, m)
     @testset "Bias type $bt" for bt in (Flux.zeros, nobias)
       m = dm(bt)
       loadparams!(m, params(m))