doc changes re at-functor and at-layer (#2390)

mcabbott · darsnack · web-flow · commit 5f84b68bf2a8 · 2024-03-06T18:46:54.000-06:00
* doc changes re at-functor and at-layer

* fix a doctest

* more fixes

* public at-layer

* add a sentence comparing to freeze/thaw

* Apply suggestions from code review

Co-authored-by: Kyle Daruwalla &lt;daruwalla@wisc.edu&gt;

* two fixes re SignDecay

---------

Co-authored-by: Kyle Daruwalla &lt;daruwalla@wisc.edu&gt;
diff --git a/NEWS.md b/NEWS.md
@@ -7,7 +7,7 @@ See also [github's page](https://github.com/FluxML/Flux.jl/releases) for a compl
   This also adds `show` methods for pretty printing.
 
 ## v0.14.12
-* New `SignDecay` optimiser, like `` WeightNorm` but for L1 norm.
+* New `SignDecay` optimiser, like `WeightDecay` but for L1 norm.
 
 ## v0.14.0 (July 2023)
 * Flux now requires julia v1.9 or later.
diff --git a/Project.toml b/Project.toml
@@ -1,6 +1,6 @@
 name = "Flux"
 uuid = "587475ba-b771-5e3f-ad9e-33799f191a9c"
-version = "0.14.12"
+version = "0.14.13"
 
 [deps]
 Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
diff --git a/docs/src/models/advanced.md b/docs/src/models/advanced.md
@@ -142,7 +142,7 @@ Join(combine, paths...) = Join(combine, paths)
 ```
 Notice that we parameterized the type of the `paths` field. This is necessary for fast Julia code; in general, `T` might be a `Tuple` or `Vector`, but we don't need to pay attention to what it specifically is. The same goes for the `combine` field.
 
-The next step is to use [`Functors.@layer`](@ref) to make our struct behave like a Flux layer. This is important so that calling `params` on a `Join` returns the underlying weight arrays on each path.
+The next step is to use [`Flux.@layer`](@ref) to make our struct behave like a Flux layer. This is important so that calling `Flux.setup` on a `Join` maps over the underlying trainable arrays on each path.
 ```julia
 Flux.@layer Join
 ```
diff --git a/docs/src/models/basics.md b/docs/src/models/basics.md
@@ -255,7 +255,7 @@ m(5) # => 26
 
 ## Layer Helpers
 
-There is still one problem with this `Affine` layer, that Flux does not know to look inside it. This means that [`Flux.train!`](@ref) won't see its parameters, nor will [`gpu`](@ref) be able to move them to your GPU. These features are enabled by the [`@functor`](@ref Functors.@functor) macro:
+There is still one problem with this `Affine` layer, that Flux does not know to look inside it. This means that [`Flux.train!`](@ref) won't see its parameters, nor will [`gpu`](@ref) be able to move them to your GPU. These features are enabled by the [`@layer`](@ref Flux.@layer) macro:
 
 ```julia
 Flux.@layer Affine
diff --git a/docs/src/models/functors.md b/docs/src/models/functors.md
@@ -2,7 +2,11 @@
 
 Flux models are deeply nested structures, and [Functors.jl](https://github.com/FluxML/Functors.jl) provides tools needed to explore such objects, apply functions to the parameters they contain, and re-build them.
 
-New layers should be annotated using the `Functors.@functor` macro. This will enable [`params`](@ref Flux.params) to see the parameters inside, and [`gpu`](@ref) to move them to the GPU.
+!!! compat "Flux ≤ 0.14"
+    All layers were previously defined with the `Functors.@functor` macro.
+    This still works, but it is recommended that you use the new [`Flux.@layer`](@ref Flux.@layer) macro instead.
+    Both allow [`Flux.setup`](@ref Flux.setup) to see the parameters inside, and [`gpu`](@ref) to move them to the GPU, but [`Flux.@layer`](@ref Flux.@layer) also overloads printing,
+    and offers a way to define `trainable` at the same time.
 
 `Functors.jl` has its own [notes on basic usage](https://fluxml.ai/Functors.jl/stable/#Basic-Usage-and-Implementation) for more details. Additionally, the [Advanced Model Building and Customisation](@ref man-advanced) page covers the use cases of `Functors` in greater details.
 
diff --git a/docs/src/models/layers.md b/docs/src/models/layers.md
@@ -12,7 +12,7 @@ The `Dense` exemplifies several features:
 
 * The bias vector is always initialised [`Flux.zeros32`](@ref). The keyword `bias=false` will turn this off, i.e. keeping the bias permanently zero.
 
-* It is annotated with [`@functor`](@ref Functors.@functor), which means that [`params`](@ref Flux.params) will see the contents, and [`gpu`](@ref Flux.gpu) will move their arrays to the GPU.
+* It is annotated with [`@layer`](@ref Flux.@layer), which means that [`Flux.setup`](@ref Flux.setup) will see the contents, and [`gpu`](@ref Flux.gpu) will move their arrays to the GPU.
 
 By contrast, `Chain` itself contains no parameters, but connects other layers together.
 The section on [dataflow layers](@ref man-dataflow-layers) introduces others like this.
diff --git a/docs/src/saving.md b/docs/src/saving.md
@@ -16,12 +16,12 @@ julia> struct MyModel
            net
        end
 
-julia> Flux.@functor MyModel
+julia> Flux.@layer MyModel
 
 julia> MyModel() = MyModel(Chain(Dense(10, 5, relu), Dense(5, 2)));
 
 julia> model = MyModel()
-MyModel(Chain(Dense(10 => 5, relu), Dense(5 => 2)))
+MyModel(Chain(Dense(10 => 5, relu), Dense(5 => 2)))  # 67 parameters
 
 julia> model_state = Flux.state(model);
 
diff --git a/docs/src/training/optimisers.md b/docs/src/training/optimisers.md
@@ -112,6 +112,7 @@ Similar to optimisers, Flux also defines some simple decays that can be used in
 ExpDecay
 InvDecay
 WeightDecay
+SignDecay
 ```
 
 ## Gradient Clipping
diff --git a/docs/src/training/training.md b/docs/src/training/training.md
@@ -384,6 +384,9 @@ Flux.thaw!(opt_state)
     The earlier "implicit" equivalent was to pass to `gradient` an object referencing only
     part of the model, such as `Flux.params(bimodel.layers.enc)`.
 
+While `adjust!` and `freeze!`/`thaw!` make temporary modifications to the optimiser state,
+permanently removing some fields of a new layer type from training is usually done
+when defining the layer, by calling for example [`@layer`](@ref Flux.@layer)` NewLayer trainable=(weight,)`.
 
 ## Implicit or Explicit?
 
diff --git a/src/Flux.jl b/src/Flux.jl
@@ -34,11 +34,11 @@ export Chain, Dense, Embedding, Maxout, SkipConnection, Parallel, PairwiseFusion
 
 @compat(public, ( # mark unexported symbols as API, on Julia 1.11
   # modules
-  Losses,
+  Losses, Train,
   # layers
   Bilinear, Scale, dropout,
   # utils
-  outputsize, state,
+  outputsize, state, create_bias, @layer,
 ))
 
 include("optimise/Optimise.jl")
diff --git a/src/functor.jl b/src/functor.jl
@@ -286,7 +286,7 @@ _paramtype(::Type{T}, x::AbstractArray{<:Complex{<:AbstractFloat}}) where {T<:Ab
     f32(m)
 
 Converts the `eltype` of model's *floating point* parameters to `Float32` (which is Flux's default).
-Recurses into structs marked with [`@functor`](@ref).
+Recurses into structs marked with [`@layer`](@ref Flux.@layer).
 
 See also [`f64`](@ref) and [`f16`](@ref).
 """
@@ -296,7 +296,7 @@ f32(m) = _paramtype(Float32, m)
     f64(m)
 
 Converts the `eltype` of model's *floating point* parameters to `Float64`.
-Recurses into structs marked with [`@functor`](@ref).
+Recurses into structs marked with [`@layer`](@ref Flux.@layer).
 
 See also [`f32`](@ref) and [`f16`](@ref).
 """
@@ -306,7 +306,7 @@ f64(m) = _paramtype(Float64, m)
     f16(m)
 
 Converts the `eltype` of model's *floating point* parameters to `Float16`.
-Recurses into structs marked with [`@functor`](@ref).
+Recurses into structs marked with [`@layer`](@ref Flux.@layer).
 
 Support for `Float16` is limited on many CPUs. Julia may
 convert to `Float32` for each operation, which is slow.
@@ -330,7 +330,7 @@ Chain(
 """
 f16(m) = _paramtype(Float16, m)
 
-# Functors for certain Julia data structures
+# Functors for certain Julia data structures -- PIRACY, should move to Functors.jl
 @functor Cholesky
 trainable(c::Cholesky) = ()
 
diff --git a/src/layers/macro.jl b/src/layers/macro.jl
@@ -7,12 +7,12 @@
 This macro replaces most uses of `@functor`. Its basic purpose is the same:
 When you define a new layer, this tells Flux to explore inside it
 to see the parameters it trains, and also to move them to the GPU, change precision, etc.
+
 Like `@functor`, this assumes your struct has the default constructor, to enable re-building.
+If you define an inner constructor (i.e. a function within the `struct` block) things may break.
 
 The keyword `trainable` allows you to limit this exploration, instead of visiting all `fieldnames(T)`.
 Note that it is never necessary to tell Flux to ignore non-array objects such as functions or sizes.
-* If some fields look like parameters but should not be trained,
-  then `trainable` lets you specify which fields to include, while the rest are ignored.
 
 The macro also handles overloads of `show` for pretty printing.
 * By default, it adds methods to 3-arg `Base.show` to treat your layer much like `Dense` or `Conv`.
@@ -21,7 +21,7 @@ The macro also handles overloads of `show` for pretty printing.
 
 (You probably still want to define 2-arg `show(io::IO, x::Layer)`, the macro does not touch this.)
 
-Note that re-running the macro with different options may not overwrite all methods, you will need to restart.
+Note that re-running the macro with different options may not remove all methods, you will need to restart.
 
 # Example
 ```jldoctest
