Well Christmas long this year… For nearly two months I’ve not done much with this project. In part that’s down to a recent computer upgrade allowing me to play games on “ultra” graphics settings, but this new RTX 2070 MaxQ 8GB GPU doesn’t really want to render Civ 6 and Forza, it wants to calculate weights for neural networks!
Document Better
I’m sure I’m not the only one who struggles to get back into a project after a break. I really enjoy it when i get going and start writing code again but this time I must have lost a couple of evenings to just opening notebooks and just having no idea where I was up to.
It was easier when I was coding computer games, I’d sketch out how it was to function and then just add to it. By comparison data projects have very little code, but so far in my journey has much more learning. And learning is what draws me back to it, like the ultimate puzzle where victory conditions get replaced with harder ones just as you complete them.
I really must document better. I’m fine with writing comments describing what the code is doing but what I must start doing is writing documentation on what I was trying to do and why I did it. A plan to remind myself what I’m doing rather than a trail of breadcrumbs through a maze of dead ends.
Everything Breaks
Another reason to document better. When you return to a project, sometimes thinks have broken. Recently a change to the NHL’s API broke one of my data scraping scripts. It was easy enough to fix but it would have been useful to have a note of what scripts are currently deployed to make testing quicker.
Similarly during my absence from this project I was getting emails from AWS reminding me to update my databases SSL certificate. I considered just turning off the database as most the bricks have been tagged now and would also save me the $0.11 its costing. I recalled I was using the same instance in my wordpress backend, fixing it thus became a priority. Thankfully AWS’s documentation is much better than mine…
On the topic of most bricks being tagged:
It is perhaps time I focused on getting a model deployed. I now have 2,092 detections over 26 images which I don’t think is enough to train a very useful model, but should not deter me from experimenting and figuring out how to deploy on the web. My hope is people will be intrigued enough to consent to allow their Lego photos to be added to the training dataset, allowing my original cyclic-improvement workflow to really get going.
