Michael Abbott

OK, re `$dst` that is good to hear. In addition to the data they have tracking information `param(rand(2)).tracker`, which Flux's `track` and `@grad` know how to update. Or rather, they...

I don't think this is supported. It's just broadcasting, `C = R .* reshape(M, 2, 1, 5)`, which if you like index notation, TensorCast.jl will write for you. For simple...

> `@tensor C[a,b,c,d] := A[a,b,c,d,e] * conj(B[a,b,c,d,e])` I believe the rule for this package is that every index must appear twice, either both on the right, or left & right....

If you comment out the last line of `f`, then its return type is `EinCode{_A,_B} where _B where _A` -- so I don't know how much hope there is of...

So this is almost precisely what my package does (with apologies for the advertising): ```julia using TensorCast, Statistics, LinearAlgebra T = randn(10,10,5); @cast E[i,n] := eigen(T[:,:,n]).values[i] # generalised mapslices p...

You can write `@ein y[] := conj(x1)[i] * x2[i]`, it turns out, or `ein"i,i->"(conj(x1), x2)`. But perhaps not ideal?

I guess you could have notation like `ein"i*,i->"(x,y) == dot(x,y) == @ein x[i]' * y[i]`, or something? BLAS stuff all has methods for conjugated / transposed / normal, but adjusting...

Isn't this example the thing which already works? Integer taken to mean non-differentiable: ``` julia> mysinpi(x) = sinpi(x) mysinpi (generic function with 1 method) julia> function ChainRulesCore.rrule(::typeof(mysinpi), x) sinpi_pullback(dy) =...

I suspect we could remove a lot of details without losing much. Has anyone ever got any value out of median GC percentage (whatever that is)? Putting the most important...

In defence of brevity, it seems that `min 18.860 μs, mean 20.375 μs` does a fair job of capturing the "about 1us wide" sentiment here. And the left tail is...