Deepcopy of Dictionaries in combination with mutating leads to errors.

[ntausend]

Hey, I discovered a strange behavior which I am not sure it is intended in that way.

Consider the following example code:

function mutating1(d, a)
	upd = d[1] * a
	d[1] = upd
	return d
end

function mutating2(d,a)
	nd = deepcopy(d)
		
	upd = d[1] * a
	nd[1] = upd
	return nd
end

function mutating3(d,a)
	nd = deepcopy(d)
		
	upd = nd[1] * a
	d[1] = upd
	return d
end

function nrmsq(d)
	res = norm(d[1]*d[2])
	return res
end

d = 3
arry = map(_ -> randn(d,d), 1:2)
dic = Dict{Any, Any}()
for (jj, tm) in enumerate(arry)
    dic[jj] = tm
end

dicco = deepcopy(dic)
f1 = x -> nrmsq(mutating1(x, 5))
f2 = x -> nrmsq(mutating2(x, 5))
f3 = x -> nrmsq(mutating3(x, 5))

# all three are equal, make sure f1 which modifies the input is called last
# same holds for f3, so make sure to call it on an copy.
f2(dic) |> display
f3(dicco) |> display
f1(dic) |> display

# works
g1 = gradient(f1, dic)[1]

# do not work
g2 = gradient(f2, dic)[1]
g3 =gradient(f3, dic)[1]

The versions utilizing f2 and f3 are leading to strange AssertionError: a === b from accum while the version with f1 works flawless.

Actually I have no idea how to deal with this and how to get an workaround.

Thanks for help in advance!

[ToucheSir]

The assertion in question: https://github.com/FluxML/Zygote.jl/blob/v0.6.40/src/lib/base.jl#L27. I don't see any reason to not support this form of mutation as well, but perhaps there are some edge cases I've not thought of.

[ntausend]

Ok thanks for the reply.

Yes i guess thats the raised assertion, but I do not understand why its raised since I also not understand how mutation is implemented.

I tried understanding the grad_mut(__contex__) appearing in all functions.

So if its possible for you to including this case it would be great, otherwise I will find an alternative to that problem :)

[CarloLucibello]

Current status of this is that the gradient doesn't throw an error anymore but g2 and g3 are different from g1:

julia> g1
Dict{Any, Any} with 2 entries:
  2 => [25.294 5.80564 -3.12013; 1.44079 -11.202 0.970101; -12.9575 7.81748 -0.0808782]
  1 => [3.31527 0.25157 -3.46752; 5.09686 -1.24741 -5.60557; -0.426827 -0.741984 0.333482]

julia> g2
Dict{Any, Any} with 2 entries:
  2 => [126.47 29.0282 -15.6007; 7.20393 -56.0099 4.8505; -64.7873 39.0874 -0.404391]
  1 => [3.31527 0.25157 -3.46752; 5.09686 -1.24741 -5.60557; -0.426827 -0.741984 0.333482]

julia> g3
Dict{Any, Any} with 2 entries:
  2 => [126.47 29.0282 -15.6007; 7.20393 -56.0099 4.8505; -64.7873 39.0874 -0.404391]
  1 => [3.31527 0.25157 -3.46752; 5.09686 -1.24741 -5.60557; -0.426827 -0.741984 0.333482]

julia> g2 == g3
true

julia> g1 == g2
false

julia> g1[1] ≈ g2[1]
true

julia> g1[2] ≈ g2[2]
false