Fix Saving and loading model output example

[logankilpatrick]

Right now, the example here: https://fluxml.ai/Flux.jl/stable/saving/#Saving-and-Loading-Models is incorrect.

julia> model = Chain(Dense(10,5,relu),Dense(5,2),softmax)
Chain(Dense(10, 5, NNlib.relu), Dense(5, 2), NNlib.softmax)

is not what happens and the loading model example at the bottom expects Chain(Dense(10, 5, NNlib.relu), Dense(5, 2), NNlib.softmax) so this fixes this inconsistency.

[ToucheSir]

What version of Flux are you on? Both of those have been reexported for as long as I can remember:

julia> using Flux

julia> softmax
softmax (generic function with 2 methods)

julia> relu
relu (generic function with 2 methods)

[logankilpatrick]

I guess I am on a very old version Flux v0.8.3 but when I do up Flux, it does not change.

[logankilpatrick]

Okay I updated to Flux v0.11.6 and still get:

julia> model = Chain(Dense(10,5,relu),Dense(5,2),softmax)
Chain(Dense(10, 5, relu), Dense(5, 2), softmax)

[logankilpatrick]

Even on the newest version, it is still wrong:

julia> using Flux

julia> model = Chain(Dense(10,5,relu),Dense(5,2),softmax)
Chain(Dense(10, 5, relu), Dense(5, 2), softmax)

(@v1.6) pkg> st
      Status `~/.julia/environments/v1.6/Project.toml`
  [fbb218c0] BSON v0.3.4
  [8f4d0f93] Conda v1.5.2
  [587475ba] Flux v0.12.7
  [5fb14364] OhMyREPL v0.5.10
  [438e738f] PyCall v1.92.3

[logankilpatrick]

But maybe I have the PR backwards, I should instead switch it to Chain(Dense(10,5,relu),Dense(5,2),softmax) in both cases?

[ToucheSir]

This is what I get:

julia> using Flux

julia> using BSON: @save, @load

julia> model = Chain(Dense(10,5,relu),Dense(5,2),softmax)
Chain(
  Dense(10, 5, relu),                   # 55 parameters
  Dense(5, 2),                          # 12 parameters
  NNlib.softmax,
)                   # Total: 4 arrays, 67 parameters, 524 bytes.

julia> @save "mymodel.bson" model

julia> model = nothing

julia> @load "mymodel.bson" model

julia> model
Chain(
  Dense(10, 5, relu),                   # 55 parameters
  Dense(5, 2),                          # 12 parameters
  NNlib.softmax,
)                   # Total: 4 arrays, 67 parameters, 524 bytes.

[logankilpatrick]

@ToucheSir exactly! The docs right now show the output of:

julia> model = Chain(Dense(10,5,relu),Dense(5,2),softmax)

as

Chain(Dense(10, 5, NNlib.relu), Dense(5, 2), NNlib.softmax)

but we both do not get that, we get:

Chain(Dense(10,5,relu),Dense(5,2),softmax)

[darsnack]

Looks good to me! This probably affects more than just the saving/loading docs? @logankilpatrick do you want to merge as is, or would you like to tackle the other cases too?

[logankilpatrick]

I'll hunt down the others when I have time (hopefully this weekend).

[darsnack]

Okay let's merge this for now then.

bors r+

[darsnack]

bors r-

[bors]

Canceled.

[darsnack]

Sorry for the spam notifications. Confused myself for a second.

bors r+

[bors]

Build succeeded:

• buildkite/flux-dot-jl