travis-leith

> The next steps are on Microsoft's side. Any idea who is "in charge" of this repo?

I found this bug by making sure that Decode.Auto.fromString works the same way as Decode.Auto.generateDecoder, if you are wondering why I use 2 different techniques here.

`chooseV` would be really useful for a numerical experiment I am working on.

If pyrsistent were to be rewritten based on dataclasses, would this be a non issue?

Is there no other way to hook into Copilot? Somehow PyCharm does it. Even RStudio has rudimentary support. Does it have to be the official Copilot extension when it comes...

@jmcphers is this the open source repo we have been waiting for? https://github.com/microsoft/vscode-copilot-chat

I suspect the solution will involve some kind of callback. The [checkpoint documentation](https://mlverse.github.io/luz/articles/checkpoints.html) seems to deal with recovering from a failure. In this case it is not a failure. Each...

@dfalbel what does the next comment (https://github.com/mlverse/luz/issues/113#issuecomment-1537450958) refer to?

This is a snippet of my attempt at achieving this, but it does not work. ```R net self$fc2() |> torch_squeeze() } ) } fitted setup( loss = function(y_hat, y_true) nnf_cross_entropy(y_hat,...

The following seems to work quite well ```R resume_training_callback # set_opt_hparams(weight_decay = 0.001) |> fit( train_ds, epochs = 5, callbacks = list(resume_training_callback("model_weights.rds")) ) ``` These callbacks are quite elegant. @dfalbel...