Interactive notebook examples

[bobturneruk]

Hi everyone,

I've been looking at how we might make interactive notebook examples (e.g. with Google Collab or myBinder) for some of the examples. It seems like training times of a couple of hours might be fairly typical (perhaps this can be sped up with better tensorflow config?). Given this, I think having notebooks for people to run online is probably not something anyone will have the patience for. I suggest that our use of notebooks might instead be around having things that people can download and run, or look at "statically" as part of online documentation.

Any thoughts on this?

All the best,

Bob

[haipinglu]

Thanks Bob. This just confirms the need of tutorials and examples are not serving the purpose well.

We can definitely make the training time bearable (in minutes rather than hours). The solution is to configure the examples to the fast and compact mode. The issue is also similar to testing.

You may start with the cifar or digits example and I can provide input on simplifying the config. We can talk about this on Thursday.

[bobturneruk]

Removing this line fixes

pykale/examples/digits_dann_lightn/main.py

Line 69 in 9b9ffcf

gpus=args.gpus,

[haipinglu]

Is there some documentation on how to populate the .yaml config files e.g. what headings, names, types are recognised? I can infer a bit from the examples... Sorry if I'm missing something obvious!

See links at the 6th bullets of https://github.com/pykale/pykale/blob/main/.github/CONTRIBUTING.md#coding-style

[haipinglu]

Removing this line fixes

Yes, tutorials should not require GPU to run. Thanks.

[bobturneruk]

Cheers @haipinglu - I feel like more docs on use of .yaml may be helpful to users. Something to discuss more, perhaps.

[haipinglu]

Cheers @haipinglu - I feel like more docs on use of .yaml may be helpful to users. Something to discuss more, perhaps.

Yes, thanks. A card for it: https://github.com/pykale/pykale/projects/9#card-61402428

[haipinglu]

Very good examples here: https://github.com/pytorchbearer/torchbearer