I have tried for 20 years to get into coding, and among adhd and having 10 million other projects going on, just could never get it beyond absolute basics and knowing some differences between languages.

Now it seems every tutorial I see is really just clicking around in a gui. Very little actual typing of code, which is the part I actually find cool and interesting.

So my question is, since everyone on lemmy is a programmer, what do you guys actually do? Is it copying and pasting tons of code? Is it fixing small bugs in Java for a website like “the drop down field isn’t loading properly on this form”?

I just dont get what “a full stack developer sufficient in sql and python” actually does. Also i dont know if that sentence even made sense!

  • dream_weasel@sh.itjust.works
    3·
    18 hours ago

    My last project is using machine learning to sort data for a company and the flow kinda goes like this:

    1. Use the advanced tab in the browser on the UI search pagr for the company to expose the API query for the data I want to have
    2. Write some shell script to programmatically call the API for the data I want.
    3. Realize that there’s actually a limit to how much data the API can return BUT the metadata says how much there is beyond the limit.
    4. Write some script to paginate (scrape over) all that data and save as a JSON file
    5. Realize my JSON is badly formatted and some queries are empty, so do some error checking and massage to make it work right.
    6. Create a python script to ingest the parts of the saved data I want and write it to a SQLite database. This takes ongoing refinement.
    7. Do some sql to count how many entries I got in the time period I was getting data for, and compare to a tool the company has to aggregate said data. It was close enough.
    8. Go learn about Jaccard similarity and locality sensitive hashing. Use this to write a script to deduplicate the database so I don’t have a zillion repeated entries. Also python, but it starts with some cursor stuff and sql.
    9. Do the actual project which is trying to get some reinforcement learning to label the data automagically. This means I also have to write some methods to get particular mixes of the data I’ve already collected. For example I want my mix to have 50 percent red balls, 20 percent green balls and 30 percent blue balls. How do you pick the right number of balls from the total set? It’s not hard but it takes some thinking.
    10. Wrap this thing up in a script that will do some load balancing on a machine with lots of gpus and CPUs so I don’t hog the entire server rack. Submit it and let it cook.
    11. Write it again so I can wrap options around it and optimize hyperparameters.
    12. Start working up the results to show improvement overall. If it’s better than what the company already has…
    13. Time to fire up the old git repo and make sure the commits are clean. Add readme and make files (bc I prefer to deploy with make) and put it in the hands of another engineer who will roll it out for that company.

    Idk it’s not a super sophisticated problem but it has a lot of moving parts and you have to kind of tackle them in order. Mostly it’s “hey here’s an obstacle to the end goal, how do I fix it on my own in a smart way?”

    Then do it.