mlr3torch

mlr3torch copied to clipboard

sebffischer

leanify nn_module

1

sebffischer

sebffischer

Lazy Tensor Extension Ideas

1

The `lazy_tensor` datatype is currently restrictive in various ways: 1. Currently there can only be a single preprocessing graph for the whole tensor-column. 1. Currently it is assumed that there...

sebffischer

It would be great if continuing of training was possible from checkpoint, currently the optimizer and loss state are not stored.

sebffischer

deep clone for PipeOpModule

1

sebffischer

Deep cloning of Learners

2

It would be nice to be able to create a proper deep clone of a trained torch learner. * [x] Support to clone `nn_modules()` has been already added in torch...

sebffischer

Checkout https://github.com/e-sensing/torchopt for more optimizers

sebffischer

should be added when there is the new paradox as this parameter will be used often.

sebffischer

Input checking must be done before entering the private `.train`

1

`mlr3torch` would benefit from some additional input checks (e.g. whether the lazy tensor shape works with a specific learner), for that the `mlr3::assert_task_learner()` method would have to call the learner's...

sebffischer