Jinja2 templating

[agitter]

We discussed writing out the Python code that reproduces the steps taken in the GUI as a desirable future feature. Jinja2 is worth considering for that. Because scikit-learn classifiers follow similar structures, we may be able to use templates to store the classifier type and arguments provided, updating a generic classification workflow template.

[agitter]

I've had some experience with Jinja2 in this Markdown-based manuscript project:

• Replacing templates with values https://github.com/greenelab/manubot/blob/27456594c407f3fed35804db4bcf3b5bba88a716/manubot/manubot.py#L343
• Example of a Markdown file with templates https://raw.githubusercontent.com/greenelab/manubot-rootstock/master/content/00.front-matter.md

[agitter]

@cmilica and I discussed using Jupyter notebooks in addition to or instead of the graphical interface, and we should continue to weigh the pros and cons. Some advantages of a Jupyter notebook are that it can be run remotely with no local software installation and it connects more directly with how scikit-learn is used for real analysis A disadvantage is that the syntax could be intimidating to participants without programming backgrounds or distract from the ML concepts.

We may want to trial some notebook-based exercises with former participants.

Generating notebooks dynamically from the graphical interface is still a possibility but would require a substantial software engineering effort.