support for alternative notebook PDF converters (e.g. webpdf)

[edublancas]

(see #672 and #658 for context)

There are a few engines to convert .ipynb to .pdf, however, we only support the default one provided by nbconvert, which depends on pandoc and text. It has a few downsides such as not being able to render embedded charts (see #658), we should add support for other engines

[neelasha23]

Can I work on this issue?

[idomic]

Go ahead! Thank you for contributing and reach out if you have any issues.

[edublancas]

A few pointers:

This is the implementation of the task that runs scripts. the nbconvert_exporter_name can switch to webpdf

And this is where the notebook conversion logic happens - we rely on nbconvert for format conversion so please check the docs.

Parameters from the pipeline.yaml are passed to the NotebookConverter.
example:

  - source: fit.py
    nbconvert_exporter_name: webpdf
    name: fit
    product:
        model: output/model.pickle
        nb: output/report.pdf

If you try this, it'll fail and ask you to run pip install nbconvert[webpdf], that's ok, the user can install it once they see the message.

If you try again, it'll complain that chromium is missing. The exporter has an option to automatically download it so we need to ensure that when we instantiate the webpdf exporter, we set that to true. So you'll need to dig into nbconvert's docs/surce code

[neelasha23]

@edublancas Thanks a lot for the pointers!

[edublancas]

(this will be part of the next release, coming up in a few days)

closing this. thanks to @neelasha23 for putting this together!

I'll create a new section in the docs to show this and a few extra things, for the moment being, this is how to use webpdf:

  - source: scripts/fit.py
    product:
      nb: products/report.pdf
    nbconvert_exporter_name: webpdf  

This will download chromium if needed, note that this requires: pip install "nbconvert[webpdf]"