Benjamin Bossan

@aliechoes I'm very happy to hear that you could make it work and it doesn't look too complicated. If you happen to create a self-contained notebook using this, I can...

Yes, just create a PR with your tutorial in it. If it's a standalone notebook, you can add it to the `notebooks` folder of skorch. for more elaborate work (e.g....

I haven't tried mixed precision training yet. Is there anything in skorch that *prevents* mixed precision training? AFAIR, we do not convert dtypes anywhere.

For me the first thing to test would be whether mixed precision training works at all or whether we have some parts in skorch that prevent that. But I don't...

First of all, this is not restricted to using the net as a transformer, even for the classical use, this could come and bite you. Furthermore, this is an issue...

So I couldn't completely understand each argument, I probably would have to implement this myself or see some code. But let me still make a proposal: Let's add a script...

Thank you for explaining the situation so well. One philosophy of skorch is that we should try to have the training, validation, and the actual prediction procedure be as close...

Overall, I like your suggestions, especially since it only requires minimal changes in `train_step` and `fit_loop`. What I would discuss is whether we want to have `criterion_` return either a...

No news on this front unfortunately. What exactly is the problem you need to solve? As mentioned in an earlier comment, a quick and dirty solution is to override `get_loss`...

This is a good question. Regarding general image augmentation, we have examples here: https://github.com/dnouri/skorch/blob/master/notebooks/Transfer_Learning.ipynb https://github.com/dnouri/skorch/blob/master/examples/nuclei_image_segmentation/Nuclei_Image_Segmentation.ipynb Regarding your use case, these examples may not fit exactly, though. We probably need a...