Carlo Lucibello

PyTorch feature parity

94

A list of PyTorch 1.7 features. Items are checked if we have something more or less equivalent in Flux or in the julia ecosystem and supported by Flux. This list...

Deprecate Flux.Optimisers and implicit parameters in favour of Optimisers.jl and explicit parameters

17

So all the things are in place and we can get rid of the current pattern using implicit params: ```julia using Flux ps = Flux.params(model) opt = Flux.Optimise.ADAM() gs =...

cleanup Flux.Losses documentation

3

afterthought of #1928

remove Bors

3

It's just extra friction with no real benefit

See https://github.com/lorenzoh/DataAugmentation.jl/issues/56

DCGAN's implementation could be optimized having just one generator's forward pass instead of two, see https://github.com/FluxML/model-zoo/pull/207#discussion_r386156978 Something like https://github.com/FluxML/Zygote.jl/pull/465 may help @matsueushi

missing Jacobian and Hessian methods

2

eventually trough Jacobian-vector product. Nice related blog post https://j-towns.github.io/2017/06/12/A-new-trick.html

missing algebra gradients

3

- [x] `diag`, `diagm` - [x] `triu`, `tril` - [x] `kron` - [x] `dot` - [x] `inv` - [x] `trace` - [x] `det` (see https://github.com/pytorch/pytorch/pull/3816) - [x] `logdet`, `logabsdet` -...

[wip] hessian, jacobian and their vector products

8

fixes #54

integrate DiffRules

1

it would be good to slim down autograd using DiffRules.jl: https://github.com/JuliaDiff/DiffRules.jl/blob/master/src/rules.jl