Carlos Parada

Ahh, I see -- I had a different bug (this one in my code) that generated a similar warning message, so I thought it was the same bug.

> Right, `@tensor` will often win, on cases they both handle, on large arrays. But not always -- sometimes it needs to call `permutedims` which is quite expensive, while `@tullio`...

@mcabbott any thoughts on the above?

> I'm surprised that fails, it's so simple. But maybe something is tripping over the type of `to_sum`? Can you post the error message, or the warning? Bizarrely, it's disappeared...

> I'm surprised that fails, it's so simple. But maybe something is tripping over the type of `to_sum`? Can you post the error message, or the warning? Oh weird, it's...

> Ideally Tullio's own check should not send LV things it can't handle. But perhaps it can be made smarter. What are the types of the arrays here? The arrays...

> > with an AxisKeys.jl `KeyedArray` wrapper around them. > > Okay, that's why. > > ```julia > julia> using LoopVectorization, AxisKeys > > julia> A = KeyedArray(rand(2,10), ([:a, :b],...

> The logic is roughly that `@tullio z := x[i] |> f` can equally well be `z = (@tullio _ := x[i]) |> f`, since it produces just one number....

> I'm not really sure what the fix is here, other than providing either an alternative backend (with StructArrays or something else) or providing an entirely different AbstractChains object which...

@cpfiffer > MCMCChains currently enforces an AxisArray backend, which is bad for all kinds of reasons: > It makes hierarchical and non-linear parameter spaces difficult to work with (try reconstructing...