Expanding My Google Suggest Jupyter Notebook to Include Stop Words
by Mike LevinSunday, February 05, 2023
Okay, it was many years into Google Photos. It’s not going to be overnight out. One key thing you need is a storage place big and secure enough to receive all your Google Photos images. I’m using a NAS with mirrored main drive and multiple backups. That should be good enough, and in the future I’ll do even better. But for now that’s an adequate receiving location so that I can start squeezing stuff off Google Photos a week at a time.
That project is actually now fairly well underway. The process started out on my laptop due to the Jupyter Notebook advantage. However since it needs to be running pretty much relentlessly 24x7 from now forward to get it done in a reasonable amount of time, I moved it over to a server. The server is actually just an LXD Linux instance running on my NAS, which is awesome. I finally have the advantage of a Kubernetes like container host at home without the tech liability. Just keep only 1 or 2 such containers running and you’ll be fine. This stuff will mainstream and become very easy as the years go by. Live with what the NAS manufacturers make easy for now.
Next is to make sure the job doesn’t get out of hand. Enumerate and expose everything about the process so you know how far along the jobs are, how long you have to wait in general, and what the process along the way is going to look like.
Interestingly, a lot of my work is going to switch from Jupyter Notebook based on my local laptop to vim text editor based in a text-based secure shell (no graphics) on the server.
Hmmm. Before I get to that, I want to refine my Google Suggest probing a bit. It’s awesome, but I’m sure there’s a better way to prompt probing. Think! Do better than shallow rote. Do intelligent middle. Forget brilliant depth for now. Tough to get there. Take a first step. What’s better than for each:
root + space * letter
Well, we can do this:
root + space + stop_word + space
I could combine both of the above, but that’s too-deep a probe. I don’t want to clobber Google Suggest. I just want to find the juicy search real estate others may have yet failed to address. Think!
Okay, done! It’s using a modified version f stopwords. Looks like it’s working really well. It will be fertile test ground for keyword clustering techniques. I already have a limit-finding KMeans approach in there right now for a first attempt at clustering Google keyword suggestions.
Let your mind go back to Google Photos again, and your work likely residing on the server. Maybe… perhaps I can get it to be Jupyter Notebook based locally, but I’d have to make the move to nbdev. It’s actually an acrobatic little maneuver, and the iron is really really hot. Maybe talk on that a bit. Show your OhAwf repo where you’re using exactly this technique, on nbdev2 already. It’s a good model.