Flux.jl: Error while running sample from docs.

[sklan]

Hello, I am new to Julia and Flux. I was running the sample code. I got the following error. I am using Julia 1.1.

julia> using Flux.Tracker


julia> f(x) = 3x^2 + 2x + 1
f (generic function with 1 method)

julia> df(x) = Tracker.gradient(f, x; nest = true)[1]
df (generic function with 1 method)

julia> df(2)
14.0 (tracked)

julia> d2f(x) = Tracker.gradient(df, x; nest = true)[1]
d2f (generic function with 1 method)

julia> d2f(2)
6.0 (tracked)

julia> f(W, b, x) = W * x + b
f (generic function with 2 methods)

julia> Tracker.gradient(f, 2, 3, 4)
(4.0 (tracked), 1.0 (tracked), 2.0 (tracked))

julia> W = param(2) 
2.0 (tracked)

julia> b = param(3) 
3.0 (tracked)

julia> f(x) = W * x + b
f (generic function with 2 methods)

julia> grads = Tracker.gradient(() -> f(4), params(W, b))
ERROR: UndefVarError: params not defined
Stacktrace:
 [1] top-level scope at none:0

[sklan]

julia> using Flux

julia> grads = Tracker.gradient(() -> f(4), params(W, b))
Grads(...)

This makes it run. But is it intended to be like this?

[MikeInnes]

You need to do using Flux, yeah. We can perhaps clarify the docs here.

[Mutaz94]

Hello, I'm facing same issue when running a model from the docs (simple linear model)

using Flux


W = rand(2, 5)
b = rand(2)

predict(x) = W*x .+ b

function loss(x, y)
    y′ = predict(x)
    sum((y.-y′).^2)
end

x, y = rand(5), rand(2)
loss(x, y)

gs = gradient(()->loss(x, y), params(W, b))

it keep give me "params not defined"

[arnaudmgh]

Seems like replacing params by Flux.params works.

gs = gradient(() -> loss(x, y), Flux.params(W, b))

[DhairyaLGandhi]

We could address it in the docs and qualify params appropriately.

[CarloLucibello]

I think params has been qualified in the docs