Write notebooks that can be used for training

[marco-c]

Either rewrite train.py and pretrain.py as notebooks, or create notebooks that import those scripts and run them.

[Shashi456]

Why a notebook ? Because it's convenient to run ?

[marco-c]

Yes, many services offer running notebooks.

[poush]

I like the idea of writing a notebook which imports these scripts and make them to run easily on notebook servers because I think it is convenient to write a script and able to test it directly through terminal.

[Shashi456]

@marco-c a few questions before writing the notebooks, so when you say rewrite them would that mean simply converting them into cells or do you want places where the code is explained as well ?
Do you expect something similar to this?

[marco-c]

@marco-c a few questions before writing the notebooks, so when you say rewrite them would that mean simply converting them into cells or do you want places where the code is explained as well ?

Simply converting them into cells, that example looks good in terms of structure.

[poush]

I have one suggestion since our primary purpose is to have a notebook file for train.py and pretrain.py. We can generate them through Travis process as part of a push to separate branch.

For eg, Commit -> Pushed to Master -> Travis Completes -> Post Travis Process Starts -> Create & Commit iPynb file and push it to separate branch say 'notebook' like 'gh-pages' concept
This way we will not be required to manage diffs on notebook file and it will be updated automatically.

For the generation process, we can use of certain type of comments to separate code into cells of the notebook. The header would be same which will load files from git after that files will be copied from train.py. We can have indicators to ignore the set of lines if required.

[marco-c]

Interesting idea! We just need to make Travis be able to push to the repo, maybe we can create a fake GitHub account and use a secret env variable on Travis to store its password.