WIP: updates for Julia 1.0 #1
Conversation
Run Travis CI against Julia 0.6
Fix deprecations
weights->w
Bump MPB bound
Fix minor typo in knapsack example
* updated struct and type parameter syntax to julia 0.6 * mutable structs, julia0.6
| @@ -205,5 +205,5 @@ function broadcast(::typeof(*), x::AbstractExpr, y::AbstractExpr) | |||
| end | |||
| broadcast(::typeof(*), x::Value, y::AbstractExpr) = DotMultiplyAtom(Constant(x), y) | |||
| broadcast(::typeof(*), x::AbstractExpr, y::Value) = DotMultiplyAtom(Constant(y), x) | |||
| broadcast(::typeof(/), x::AbstractExpr, y::Value) = DotMultiplyAtom(Constant(1./y), x) | |||
| broadcast(::typeof(/), x::AbstractExpr, y::Value) = DotMultiplyAtom(Constant(1 ./ y), x) | |||
There was a problem hiding this comment.
I believe instead of overloading Base.broadcast directly, we're now supposed to go through a different set of machinery laid out in https://docs.julialang.org/en/v1/manual/interfaces/#man-interfaces-broadcasting-1. For now this is probably okay, since it does technically work to do it this way, but I can look into reworking this later.
src/atoms/affine/stack.jl
Outdated
|
|
||
| vcat(args::AbstractExpr...) = transpose(HcatAtom(map(transpose, args)...)) | ||
| vcat(args::AbstractExprOrValue...) = transpose(HcatAtom(map(arg -> transpose(convert(AbstractExpr, arg)), args)...)) | ||
| vcat(args::Value...) = Base.cat(args..., dims=1) # Note: this makes general vcat slower for anyone using Convex... |
There was a problem hiding this comment.
Why is this definition necessary? Value is defined as Union{Number, AbstractArray}, which should already work as expected with the vcat methods in Base.
There was a problem hiding this comment.
If I comment out this line I get some strange error messages for some tests. Very likely the problem is that in this case the vcat(args::AbstractExprOrValue...) function is called for arguments, which all have Value types but seemingly this function works only for arguments that are either all AbstractExpr types or mixed AbstractExpr and Value types.
There was a problem hiding this comment.
Ah I see, it's because the fallback method that vcat([3], 3), for example, uses is untyped. This is good as is except it should use Val like the Base definition:
vcat(X...) = cat(X...; dims=Val(1))
src/atoms/affine/stack.jl
Outdated
| Base.vect(args::AbstractExpr...) = transpose(HcatAtom(map(transpose, args)...)) | ||
| Base.vect(args::AbstractExprOrValue...) = transpose(HcatAtom(map(arg -> transpose(convert(AbstractExpr, arg)), args)...)) | ||
|
|
||
| function Base.vect(args::Value...) |
There was a problem hiding this comment.
This similarly seems unnecessary, as the Base methods work.
| x = Variable(10) | ||
| a = rand(10, 1) | ||
| a = shuffle(collect(0.1:0.1:1.0)) |
There was a problem hiding this comment.
Using rand led to test failures with SCS solver for some cases. I think it happened because for some set of random numbers the problem might not be well conditioned (i.e. some numbers are too close to each other). So I changed it to a set of numbers from an equidistant grid but to still put some randomness there I shuffled them.
There was a problem hiding this comment.
Alright.
You should probably change it to 0.0:0.1:0.9 though as that is within the range of rand while 1.0 is not.
There was a problem hiding this comment.
For these tests the actual numbers don't really matter and they even don't need to be strictly within the range [0, 1). If possible I would rather not touch them as they finally seem to be passing nicely.
Updating
Convex.jlfor Julia 1.0.