Hi @thautwarm ,
In a new version of Soss, I'm trying to really nail down type stability. Using GG I was having lots of trouble getting constructions like
For(n -> Normal(α + β * x[n], σ), 1:N)
to be type-stable.
After lots of exploring, I realized this could be solved by putting this function outside of GG:
f(ctx, n) = Normal(ctx.α + ctx.β * ctx.x[n], ctx.σ)
f(ctx::NamedTuple) = Base.Fix1(f, ctx)
Then within GG, it could be called as
That works great, but it's tricky to automate, since the f needs to be made available separately within the module, and I'd rather not use eval to put it there.
I eventually found this to work:
For(let f = ctx -> Base.Fix1(ctx) do ctx, n
Normal(ctx.α + ctx.β * ctx.x[n], ctx.σ)
end
f((; α, β, σ, x))
end, 1:N)So now, I think I can just rewrite things this way before calling GG. But I wanted to check with you before doing this, since it seems plausible the current Closure approach could be updated to something similar, and maybe it would help other cases with type stability. What do you think?
Hi @thautwarm ,
In a new version of Soss, I'm trying to really nail down type stability. Using GG I was having lots of trouble getting constructions like
to be type-stable.
After lots of exploring, I realized this could be solved by putting this function outside of GG:
Then within GG, it could be called as
That works great, but it's tricky to automate, since the
fneeds to be made available separately within the module, and I'd rather not useevalto put it there.I eventually found this to work:
So now, I think I can just rewrite things this way before calling GG. But I wanted to check with you before doing this, since it seems plausible the current Closure approach could be updated to something similar, and maybe it would help other cases with type stability. What do you think?