.jl

using Flux
using Flux: @epochs
using BSON: @save
using BSON: @load
using Random
# using Flux

print("Running!")
Random.seed!(1234);

model = Dense(2, 1)

loss(x, y) = Flux.mse(model(x), y)
# @load "mymodel.bson" weights
# Flux.loadparams!(model, weights)
ps = Flux.params(model)

data = []
gen1 = () -> return rand(1:55555)

for i in 1:9000
    s = gen1()
    a = gen1()
    push!(data, ([s, a], s + a))
end

opt = ADAM()

epoch_count = 0

s = function()
    global epoch_count
    epoch_count = epoch_count + 1
    if (epoch_count == 100)
        x1 = gen1()
        x2 = gen1()
        y = x1 + x2

        println(y)
        @show(model([x1, x2]))
        epoch_count = 0
    end
end

@epochs 500 Flux.train!(loss, ps, data, opt, cb = () -> s())
# end

@save "mymodel.bson" model