Get rid of documentation warnings and 404 pages (#1987)

* Get rid of documentation warnings

* fix cross-references

* add missing docstrings

* Use `onehotbatch` instead of internal struct

* Fix doctests

* Add `Optimiser.jl` as a doc dependency

Add `cpu` and `gpu` to the manual and `Optimisers.jl` as a dependency

Author: Saransh-cpp

docs/Project.toml

@@ -4,6 +4,7 @@ Documenter = "e30172f5-a6a5-5a46-863b-614d45cd2de4"
 Functors = "d9f16b24-f501-4c13-a1f2-28368ffc5196"
 MLUtils = "f1d291b0-491e-4a28-83b9-f70985020b54"
 NNlib = "872c559c-99b0-510c-b3b7-b6c96a88d5cd"
+Optimisers = "3bd65402-5787-11e9-1adc-39752487f4e2"
 
 [compat]
 Documenter = "0.26"

docs/make.jl

@@ -1,10 +1,10 @@
-using Documenter, Flux, NNlib, Functors, MLUtils, BSON
+using Documenter, Flux, NNlib, Functors, MLUtils, BSON, Optimisers
 
 
 DocMeta.setdocmeta!(Flux, :DocTestSetup, :(using Flux); recursive = true)
 
 makedocs(
-    modules = [Flux, NNlib, Functors, MLUtils, BSON],
+    modules = [Flux, NNlib, Functors, MLUtils, BSON, Optimisers],
     doctest = false,
     sitename = "Flux",
     pages = [

docs/src/data/mlutils.md

@@ -20,10 +20,15 @@ Below is a non-exhaustive list of such utility functions.
 
 ```@docs
 MLUtils.unsqueeze
+MLUtils.flatten
 MLUtils.stack
 MLUtils.unstack
+MLUtils.numobs
+MLUtils.getobs
+MLUtils.getobs!
 MLUtils.chunk
 MLUtils.group_counts
+MLUtils.group_indices
 MLUtils.batch
 MLUtils.unbatch
 MLUtils.batchseq

docs/src/gpu.md

@@ -86,6 +86,11 @@ julia> x |> cpu
  0.7766742
 ```
 
+```@docs
+cpu
+gpu
+```
+
 ## Common GPU Workflows
 
 Some of the common workflows involving the use of GPUs are presented below.

docs/src/models/functors.md

@@ -9,5 +9,7 @@ Functors.isleaf
 Functors.children
 Functors.fcollect
 Functors.functor
+Functors.@functor
 Functors.fmap
+Functors.fmapstructure
 ```

docs/src/models/nnlib.md

@@ -11,7 +11,9 @@ NNlib.celu
 NNlib.elu
 NNlib.gelu
 NNlib.hardsigmoid
+NNlib.sigmoid_fast
 NNlib.hardtanh
+NNlib.tanh_fast
 NNlib.leakyrelu
 NNlib.lisht
 NNlib.logcosh

docs/src/training/optimisers.md

@@ -1,3 +1,7 @@
+```@meta
+CurrentModule = Flux
+```
+
 # Optimisers
 
 Consider a [simple linear regression](../models/basics.md). We create some dummy data, calculate a loss, and backpropagate to calculate gradients for the parameters `W` and `b`.
@@ -189,3 +193,13 @@ opt = Optimiser(ClipValue(1e-3), Adam(1e-3))
 ClipValue
 ClipNorm
 ```
+
+# Optimisers.jl
+
+Flux re-exports some utility functions from [`Optimisers.jl`](https://github.com/FluxML/Optimisers.jl)
+and the complete `Optimisers` package under the `Flux.Optimisers` namespace.
+
+```@docs
+Optimisers.destructure
+Optimisers.trainable
+```

docs/src/utilities.md

@@ -59,7 +59,7 @@ Flux.f32
 
 Flux provides some utility functions to help you generate models in an automated fashion.
 
-[`outputsize`](@ref) enables you to calculate the output sizes of layers like [`Conv`](@ref)
+[`Flux.outputsize`](@ref) enables you to calculate the output sizes of layers like [`Conv`](@ref)
 when applied to input samples of a given size. This is achieved by passing a "dummy" array into
 the model that preserves size information without running any computation.
 `outputsize(f, inputsize)` works for all layers (including custom layers) out of the box.
@@ -107,7 +107,6 @@ Flux.outputsize
 
 ```@docs
 Flux.modules
-Flux.destructure
 Flux.nfan
 ```
 

src/functor.jl

@@ -54,7 +54,7 @@ Given a model or specific layers from a model, create a `Params` object pointing
 
 This can be used with the `gradient` function, see [Taking Gradients](@ref), or as input to the [`Flux.train!`](@ref Flux.train!) function.
 
-The behaviour of `params` on custom types can be customized using [`Functor.@functor`](@ref) or [`Flux.trainable`](@ref).
+The behaviour of `params` on custom types can be customized using [`Functors.@functor`](@ref) or [`Flux.trainable`](@ref).
 
 # Examples
 ```jldoctest

src/layers/basic.jl

@@ -124,7 +124,7 @@ The out `y` will be a vector  of length `out`, or a batch with
 
 Keyword `bias=false` will switch off trainable bias for the layer.
 The initialisation of the weight matrix is `W = init(out, in)`, calling the function
-given to keyword `init`, with default [`glorot_uniform`](@doc Flux.glorot_uniform).
+given to keyword `init`, with default [`glorot_uniform`](@ref Flux.glorot_uniform).
 The weight matrix and/or the bias vector (of length `out`) may also be provided explicitly.
 
 # Examples
@@ -262,7 +262,7 @@ which constructs them, and the number to construct.
 
 Maxout over linear dense layers satisfies the univeral approximation theorem.
 See Goodfellow, Warde-Farley, Mirza, Courville & Bengio "Maxout Networks" 
-[https://arxiv.org/abs/1302.4389](1302.4389).
+[https://arxiv.org/abs/1302.4389](https://arxiv.org/abs/1302.4389).
 
 See also [`Parallel`](@ref) to reduce with other operators.
 
@@ -651,7 +651,7 @@ for a vocabulary of size `in`.
 
 This layer is often used to store word embeddings and retrieve them using indices. 
 The input to the layer can be either a vector of indexes
-or the corresponding [onehot encoding](@ref Flux.OneHotArray). 
+or the corresponding [`onehot encoding`](@ref Flux.onehotbatch). 
 
 # Examples
 ```jldoctest
@@ -662,8 +662,8 @@ Embedding(1000 => 4)  # 4_000 parameters
 
 julia> vocab_idxs = [1, 722, 53, 220, 3];
 
-julia> x = Flux.OneHotMatrix(vocab_idxs, vocab_size); summary(x)
-"1000×5 OneHotMatrix(::Vector{Int64}) with eltype Bool"
+julia> x = Flux.onehotbatch(vocab_idxs, 1:vocab_size); summary(x)
+"1000×5 OneHotMatrix(::Vector{UInt32}) with eltype Bool"
 
 julia> model(x) |> summary
 "4×5 Matrix{Float32}"