Mike J Innes
Mike J Innes
If we go full steam ahead with more functional-style AD (#628), then we'll need to rework optimisers a little. I think at the core the update step will look very...
We expose a pullback/vjp-based API for gradients (`y, back = forward(f, x); x̄ = back(ȳ)`), with `x̄ = gradient(f, x)` as simple syntax sugar of top of this. This interface...
With #59 we have good infrastructure for generating notebooks. There are a couple more steps we need: * Content. This could be taken from existing open-source tutorials (#68) and we...
Stheno is great and I'm enjoying it a lot so far. In some cases I've found that it's easy to run into numerical issues when getting the `logpdf`, when the...
JET.jl rightly notes that this function will error if a `missing` appears inside an expression, and we try to bind it. This is certainly an edge case (and clearly no-one's...
Lines beginning with `]` could be considered "Pkg mode" and evaluated as if in that repl mode; this follows what IJulia does when this is in a cell. Currently this...
```julia D(f, x) = ForwardDiff.derivative(f, x) D(1) do x f(y) = (x = x*y) D(f, 1) D(f, 1) end # => 2 ``` The correct answer is 1. This is...
We already have support for this via `svd(X).S`, but support for `svdvals` directly would be nice. It'd also be great to support GenericLinearAlgebra, which I don't think we do at...
This is just a prototype for the time being (Mjolnir is not registered yet) but it shows off the basic ideas. This is the right solution to #16 (handling dispatch...
Consistent with the behaviour of the GPU stack, and others.