I needed the help to code it, since it is written in python in which I have no experience and the experiment was also more to see how useful it is to have AI support while coding.
I mean, I didn’t want to read the data manually (3100 player information), so to automate that and the calculation, I wrote the script.
For an expert in python it would have been time wasting, ending up spending 30 minutes which would have been 15 minutes by doing it manually. For someone who can’t code in any language at all, it couldn’t have done the job, so wouldn’t have been a help. But for people like me, who have some coding knowledge, it saved definitely a lot of time.
Alone the knowledge around syntax and which possibilities the language has based on libraries. If I would have needed to ask a common search engine, it wouldn’t have been a 30 minute job but 120+ minutes. With the support all I needed to do was to ask how I can achieve this and that, got examples and only had to modify it for my needs.
Often enough it was literally just copy and paste as it already was exactly what I was looking for.
I didn’t want to write a dissertation with that or publish it in the next mathematical science magazine.
But I also think you are wrong in that regard. Balance and tier lists should just show of how specs are looking compared to each other. So I just took the two most obvious parameter, combine them to one value and looked which spec is the median. From there I compared the other specs to it.
Its not rocket science but good enough for an objective overview imo!
ChatGPT is usually very good and rephrasing or paraphrasing something but it’s bad at actual thinking. For coding it often gives a good starting point to go from and it’s also really good at explaining frameworks to you while it’s really bad at problem sovling, for example leetcode questions (most programmers can’t solve these either tho to be fair ).
When you can google a solution to your problem when it’s quite likely that ChatGPT can do that too (unless it’s past the trainings data cutoff). I kinda think of it as a search engine on steroids.
The list itself looks pretty solid even tho I think that it’s not fully up to date (also need to take the people into consideration who stop playing a class once it gets nerfed but still sit on the rating). Definitely has much more substance than the usual tier lists which often times feel made up to large extends.
ye i know i worked with the battle net API before too.
100-200 rating higher is the highest player since mmr injection i think.
dh really struggles into sub rogues. Hunters are also quite okay into DHs.
Before DH was extremely strong and tanky into casters but right its pretty even. Overall DH can win everything but also lose everything which gets exploited really hard at higher ratings.
I actually didn’t work with the api. I wanted to at first but even the test button from blizzard wasn’t working. So I used the beautifulsoup library from python to basically scrap the data from the website.
They also lose everything low rating while you painfully watch hunter and mages just walk | blink | disengage out of your long darkness while you beg them and cry as they die
I am joking and have loads to improve on, this still pains me so bad though
Scrolled down to PvP Season API to get the leaderboard, used my data, pressed “Try it” but it threw some error messages that it isn’t available at the time.
ah okay weird. Should work because check-pvp and co are working too? So the actual endpoints should work but maybe the test function on the website doesnt…
although that does not make much sense haha
One good thing about this approach is that it focuses on actual winrates and some sort of activity instead of what people have been monitoring for the last decades : static spec populations.
).