Question: If I have a Lux model with a single input, how I I create one with two inputs

[TorkelE]

Basically, I have a Lux model like:

model = Lux.Chain(
    Lux.Dense(1 => 4, Lux.softplus),
    Lux.Dense(4 => 4, Lux.softplus),
    Lux.Dense(4 => 4, Lux.softplus),
    Lux.Dense(4 => 1, Lux.softplus)
)

and I like to generate a Lux model from it

model = Lux.Chain(
    Lux.Dense(2 => 4, Lux.softplus), # Changed 1 to 2
    Lux.Dense(4 => 4, Lux.softplus),
    Lux.Dense(4 => 4, Lux.softplus),
    Lux.Dense(4 => 1, Lux.softplus)
)

is there an easy way to do this?

I could of course declare it directly. However, I have a workflow for fitting functions for Lux models. Now, sometimes my functions are periodic, and I can encode this through some transfomration that reuqires to network to have two inputs. And here I need to detect it and generate the appropriate model. So Ideally I'd like to create a function like:

function make_input_2(model_1ins)
     model_2ins = deepcopy(model_1ins)
     model_2ins.later_1.in_dims = 2
     return model_2ins
end

however, I am usure of the best way to do this.

[TorkelE]

I think I have a workflow which seems to work

using Lux
model_1in = Lux.Chain(
    Lux.Dense(1 => 4, Lux.softplus),
    Lux.Dense(4 => 4, Lux.softplus),
    Lux.Dense(4 => 4, Lux.softplus),
    Lux.Dense(4 => 1, Lux.softplus)
)
model_2in = Lux.Chain(
    Lux.Dense(2 => 4, Lux.softplus), # Changed 1 to 2
    Lux.Dense(4 => 4, Lux.softplus),
    Lux.Dense(4 => 4, Lux.softplus),
    Lux.Dense(4 => 1, Lux.softplus)
)

using Setfield
model_1in_to_2in = @set model_1in.layers.layer_1.in_dims = 2

model_2in == model_1in_to_2in # true

However, my knowledge about how Lux actually work is limited, so not sure if this is appropriate (in simple tests it seems ok though)

[avik-pal]

I will strongly recommend against doing

model_1in_to_2in = @set model_1in.layers.layer_1.in_dims = 2

What is the issue with reconstructing the full model again?