Mike Levin SEO

Future-proof your technology-skills with Linux, Python, vim & git... and me!

← Concurrent Web Crawling in Python
Jupyter Notebooks Are Dev Tools With nbdev →

Okay, I'm Finally Keeping Audiences In Mind

by Mike Levin

Thursday, April 07, 2022

Solve things in the order and way they need to be solved. Don’t over-complicate things. Make a journal entry that’s appropriate for public consumption on these very issues. Yes! The vision is that every day at least one of my journal entries is made with a public audience in mind and the publication thereof is really just copy/pasting between vim buffers and doing a few git commits and pushes. THIS is the process I should start with. Pick a site, any site…

I’m not sure about that WhatsaMetaFor.io site. It’s one of those fancy top-level domains that they can easily hike the price on you every year. Nonetheless, it may be the perfect one for these types of “meta” entries. What’s a meta for? Well, it’s for self-awareness, reflection and entering a more deliberate “human” and less “animal” state.

Okay, so WhatsaMetaFor.io is for open-ended rambling like this… free association from which all the meta-observations that make all the difference arise. It’s too bad that Facebook latched onto the concept of meta and ruined its real meaning for the rest of the world. No worries. The ties to the field of SEO with “meta” tags is still enough for the cleverness of this site name to remain relevant. Add to that the actual metaphors I’ll be writing about on this site and it’s very appropriate.

The same thing; to help you think better. In the former, there is self-reflective, self-descriptive, self-aware (what have you) “extra” data layered into a system. I the case of SEO it’s invisible meta tags in the head element of an HTML file. In the case of human consciousness, it’s I think therefore I am. That’s the first meta. Both are about self-awareness.

A metaphor is a story, situation or image where things stand for something else so that the same ideas, principles or insights can be carried over to the new thing. New ideas are difficult. They are inaccessible because of quite how much is new when tackling a new topic. Metaphors create the bridge between older, familiar ideas and the new. For example the field of SEO is full of metaphors regarding link juice and the distribution of liquid authority as it flows from link to link.

That’s all well-and-good, but what about the here-and-now? I’m doing this writing so that I can struggle my way to the next and the next and the next steps. So most everything now is about the pip-installable mlseo package. It is once again my own personal grab-bag of secret weapons and continuation of making myself relevant in the now. Pipulate and Levinux are nice but old and hard to update to the new world. Maybe keep them around for posterity, but all new effort goes into mlseo. Probably one day I’ll just make Pipulate and alias to mlseo. Okay, so next step? 1, 2, 3… 1?

Nothing improves until you can solve your nbdev problem in which the documentation is not really built well with nbdev_build_docs because its default behavior is to convert every .ipynb file… wait, is this still appropriate for this site? Forge ahead and copy/paste the parts that are appropriate for each site into the appropriate site. Don’t be artificially limited based on your perception of some unlikely theoretical audience. You are doing this writing primarily for yourself and primarily in your “normal” everyday personal journal.

Okay, so that nbdev problem. Get into these things deeper. Use the metaphors. You’re on that site, dammit. Okay, so let’s talk about:

Gotta get work done. Can’t get overly-distracted by the wrong things. But it’s okay to get a little distracted by the right things. So how can you know the difference? You stand over the edge of the rabbit hole and peer in. You may in fact tie yourself to a tree with a rope and lower yourself down over the edge to get a better look around. But don’t start falling like Alice or you may never come back up and out in time to finish and deliver the original work.

The Catch+22 is that your very career and abilities move forward based on occasionally chasing the correct rabbit down the correct hole, and you just have to get a feel for which is which, or maybe be lucky enough consecutive times to be on a great path. Usually it requires a very deliberate, very meta-like process to pull off rabbit-hole navigation. They most often act as show-stoppers, pitfalls, red-herrings or what-have-you.

And so here, nbdev itself is a rabbit hole. It reinforces my decision that JupyterLab standalone is a good idea. However there’s these little gotcha’s that show me nbdev is still bleeding edge. When incorporating an async function into 00_core.ipynb, I bleed. Do I even engage the github community? Okay, sure why not. It could help me think through it. Focus and put something out there. Let’s do it in an existing open issue.

Corner this issue. Document it well. Get it out there on github without creating a duplicate issue. Include barebones-to-reproduce screenshots. Ugh, to do this in a new folder I have to initialize it as an nbdev project. Do it but just borrow the already vetted settings.ini from another folder and make changes to it.

Okay, nbdev_new doesn’t work unless it’s already a git repo. Github too? Find out…okay, git init wasn’t enough. It needs a remote.origin.url, which in itself is an issue. It shows me that nbdev isn’t really an offline tool from scratch. You need Internet at least enough to get such a new repo in github. Do it! Okay, so…

Made new public repo in Github called miklevin/issues
git branch -M main
git remote add origin git@github.com:miklevin/issues.git

Didn’t even push yet.

nbdev_new

Success! Try nbdev_build_docs right off the bat… Okay, done. It is at [Github]https://github.com/fastai/nbdev/issues/327 which is really amazing for me. I usually just suffer with such things waiting for them to work themselves out, but I can play an active role really cornering them into seeing the problem with good documentation and screenshots. Whoop!

Okay, next? Come up with an 80/20-rule workaround. Honestly, I’m really only in it for the README.md that it creates from the index.html and a series of other notebooks that I will very deliberately be adding, starting yesterday with google_analytics.ipynb

Wow, alright. That was pretty extreme. I actually made a new repo in Github and updated the README.md to link-to and document the open nbdev case that I chimed in on. Hopefully I got the right one.

Okay, back to 80/20-rule solutions. As much as I hate it, I need my own make process. I’ll probably just start with a unix script. I see no reason for anything more fancy.

Think! vim release.sh… shouldn’t be too hard! No, I need a prerelease.sh because I don’t really want to be pushing to pypy all the time. In fact, maybe it should be prepare.sh because I’ll still want to manually make pypi. Yes!