flatten layers appear as closures

[adrhill]

When using Metalhead models, layers using flatten appear as closures, e.g. Metalhead.var"#44#45"() in the following example.

This makes it impossible to dispatch on these layers or infer what they are without looking into the Metalhead source.

Example

julia> model = VGG19();

julia> model.layers
Chain(
  Conv((3, 3), 3 => 64, relu, pad=1),   # 1_792 parameters
  Conv((3, 3), 64 => 64, relu, pad=1),  # 36_928 parameters
  MaxPool((2, 2)),
  Conv((3, 3), 64 => 128, relu, pad=1),  # 73_856 parameters
  Conv((3, 3), 128 => 128, relu, pad=1),  # 147_584 parameters
  MaxPool((2, 2)),
  Conv((3, 3), 128 => 256, relu, pad=1),  # 295_168 parameters
  Conv((3, 3), 256 => 256, relu, pad=1),  # 590_080 parameters
  Conv((3, 3), 256 => 256, relu, pad=1),  # 590_080 parameters
  Conv((3, 3), 256 => 256, relu, pad=1),  # 590_080 parameters
  MaxPool((2, 2)),
  Conv((3, 3), 256 => 512, relu, pad=1),  # 1_180_160 parameters
  Conv((3, 3), 512 => 512, relu, pad=1),  # 2_359_808 parameters
  Conv((3, 3), 512 => 512, relu, pad=1),  # 2_359_808 parameters
  Conv((3, 3), 512 => 512, relu, pad=1),  # 2_359_808 parameters
  MaxPool((2, 2)),
  Conv((3, 3), 512 => 512, relu, pad=1),  # 2_359_808 parameters
  Conv((3, 3), 512 => 512, relu, pad=1),  # 2_359_808 parameters
  Conv((3, 3), 512 => 512, relu, pad=1),  # 2_359_808 parameters
  Conv((3, 3), 512 => 512, relu, pad=1),  # 2_359_808 parameters
  MaxPool((2, 2)),
  Metalhead.var"#44#45"(),
  Dense(25088, 4096, relu),             # 102_764_544 parameters
  Dropout(0.5),
  Dense(4096, 4096, relu),              # 16_781_312 parameters
  Dropout(0.5),
  Dense(4096, 1000),                    # 4_097_000 parameters
  NNlib.softmax,
)                   # Total: 38 arrays, 143_667_240 parameters, 548.053 MiB.

(jl_qLymP0) pkg> status Metalhead
      Status `/private/var/folders/74/wcz8c9qs5dzc8wgkk7839k5c0000gn/T/jl_qLymP0/Project.toml`
  [dbeba491] Metalhead v0.5.3

[adrhill]

I'm not sure whether this is just a missing reexport by Metalheads or whether I should make a PR to Flux.

I guess the introduction of a struct would be most useful for my personal use-case of dispatching on individual layer types:

struct Flatten end
(::Flatten)(x) = flatten(x)

[ToucheSir]

This should be fixed on master as of #70. You should see then flatten instead of an anonymous function.