Kaushik's Blog

The Moderately Surprising Effectiveness of Note-Taking

Today, a coworker asked me if I'd ever encountered a strange issue with a locally built Docker image not containing some TensorFlow model files. It sounded familiar to me, like something I'd dealt with some months ago.

My vague recollection was that this had something to do with Git LFS, so I punched that into the search bar. Searching a bunch of text files is blazing fast, I got 12 results for "LFS", of which the eighth was the one I wanted.

Thankfully, on January 31st, I'd jotted down the terminal output describing the problem, the exact steps I took to solve the problem, and even the StackOverflow link where I'd found the solution. Not only was I able to quickly help out my teammate, I could give him exactly the information he needed - the exact lines of code to run.

Did I have any idea, when I wrote that short note down, that it'd come in handy down the line? Absolutely not. I'd just formed a habit of writing down problems I'd faced, what they looked like, and how I overcame them.

I read somewhere that most of your notes will never be read again. I think I even made a note about it.

To some, that may seem discouraging. What's the point of doing something if you'll never use it?

I'm more inclined to the process mindset. If you stick to the process, if you trust the process, the results will make themselves seen when you really need it. Today was a nice little affirmation of that sentiment.


For the curious, here is the note in question:

root@farm-robots:/opt# cat model_data/detectron2/maskrcnn_xception.pth
version https://git-lfs.github.com/spec/v1
oid sha256:9ae08986ed83a2885c6493f8deaa130ff7bbe4ac1b4a8595f6ab8174b75555ac
size 856189863