Toby Dylan Hocking
Toby Dylan Hocking
For a test case I would suggest to modify the example code from https://github.com/tdhock/mlr3torchAUM/blob/main/man/batch_sampler_stratified.Rd First I installed mlr3torch from this PR branch, then I ran the example code. ```r devtools::install_github("mlr-org/mlr3torch@dont-require-batch-size")...
I propose the following test cases ```r sonar_task
> I still think the `batch_size` parameter should be required for prediction. why? batch size is a concept which is defined for training (= number of samples used to compute...
Also in the example code above I provided, it appears to be missing a pipeop between nn_linear and torch_loss. Why isn't this an error? I would expect some error like...
> nn_head is basically just a nn_linear that infers the output dimension from the task, but its equally valid to do create own nn_linear, so i don’t think this should...
thanks for the links to the other docs about lazy tensors. I have not had a chance to review them yet, but they seem to be helpful.
related to #420 and may be fixed by https://github.com/mlr-org/mlr3torch/pull/425
also for the video can you please narrate next time? use your voice to explain what you are doing, what you observe after executing each command, etc.
please grep tests/testthat/test-renderer-* files for "clickID" to see examples.
this looks like a reasonable start. let's see if it works...