Closed nicole-brewer closed 1 year ago
I discovered that there is actually a lot of flexibility in the naming because it doesn't rely on directories being correct. That way, I can organize the notebooks any way I want and sort them out in the library totally differently. so I will have jupyter_web_app_template.data.model
and jupyter_web_app_template.data.views.plot_view
.
... well wait. Can I just do jupyter_web_app_template.data.views import PlotView
. That is where there are multiple notebooks contributing to a single "views" module?
I'm not getting an error, but the code from data.ipynb labeled for export to data.views
is not showing up.
This makes me think I'm supposed to be able to export outside the default (as the the word "default") but I don't see any evidence of this in the documentation. My next thought is that the dependency order matters. That is, the notebook that has the (non-default) export to data.views
, may be converted first and then overwritten by the notebook with the default export to data.views
. Maybe if the dependency was listed this wouldn't happen?
I didn't get around to trying that because I'm pretty sure it wouldn't do any good. I tried to two notebooks with the same default export to see if they would combine but they don't. So I think notebooks are one-to-one with modules. Changing this would be a great feature to nbdev I think.
So here's what I've come up with. The module jupyter_web_app_template.data.view.table
will live in nbs/data/view.table.ipynb
. That way there is some directory structure but not too many levels without things in them. But honestly, for the tutorial, it may make more sense to be in a flat directory using filenames with dot notation. This way I can number all the ones to be accessed and there isn't confusion hopping in and out of directories.
So for the tutorial I will have a flat structure but if I turn it into a template it will have some directories in there for better organization.
See above comment
In the Superpower project, I split up a particular StatsModel by model and view. For example, I chose to split ANOVA into
superpower.model.ANOVAModel
andsuperpower.view.ANOVAView
. I didn't love that arrangement because I couldn't edit models and views in the same notebook.So for this project, I'm going to start by putting the Model and View in the same module, for example
jupyter-web-app-template.data.DataModel
andjupyter-web-app-template.data.DataView
. That way the first view can be developed in the same notebook as the model. Not sure what that will look like as I add more views. Can I have a default export that references another module? Or would I have to offload it tojupyter-web-app-template.data.views.DataView
? And then the original view (jupyter-web-app-template.data.DataView
) would be in a different place than all the other views which would be confusing.Maybe the answer is to create a
jupyter-web-app-template.data.views
before creating the view injupyter-web-app-template.data.ipynb
. This works as long as the models are "one-to-many" views. If there is a view that accommodates more than one kind of model, we probably have to resort the the superpower setup mentioned earlier. But I don't even think that superpower was "many-to-many" so that probably isn't an issue.