define requirements for notebooks

[daMichaelB]

🐛 Bug

CIFAR-10 Baseline Example Notebook is failing on importing torchvision as it is not part of the requirements files.

To Reproduce

Steps to reproduce the behaviour:

1. Go to https://mybinder.org
2. mybinder will start a fresh docker container with this repository and all requirements installed
3. Start the Cifar-10-baseline example.
4. torchvision fails to be imported

Expected behavior

Add torchvision to the requirements/default.txt

[Borda]

requirements are set for each notebook meta-file, so for the CIFAR it is here https://github.com/PyTorchLightning/lightning-tutorials/blob/79e1490f5f86d0aa127557fa5e660c73f88fc69d/starters/cifar10-baseline/.meta.yml#L8-L10
otherwise, the main requirements will be too heavy even for simple examples or what is your suggestion?

[Borda]

3. Start the Cifar-10-baseline example.

are you opening the python file or ipynb in .notebooks folder?

[Borda]

ok, I may found a middle ground, we can add an extra cell at beginning of each notebook which would pip install special requirements for the particular notebook :]

[daMichaelB]

Hey thank you for your fast feedback! i was using the .py file and was not aware of the .meta.yml. I just saw that you create the .ipynb files from .py, which is cool stuff! Thanks for your explanation and can also be closed from my side.

[Borda]

I think that you brought good point and I will add the extra cell, so if you use standard env you shall have it all installed and no delay... btw, do you know if mybinder.org can use your own/custom image? (thinking about #23)

[daMichaelB]

Ok i think that might be helpful to have that cell on top, thanks man! There is such a thing on mybinder.org (https://mybinder.readthedocs.io/en/latest/tutorials/dockerfile.html) but i have not tested it on my own. I only want to point out that mybinder was really helpful to debug through the code within seconds - But the downside is there is no GPU.