Hearted Youtube comments on Fireship (@Fireship) channel.
-
46000
-
36000
-
23000
-
17000
-
16000
-
13000
-
12000
-
11000
-
10000
-
10000
-
10000
-
9400
-
9400
-
7700
-
I read that guy's paper and his research is so bad! He tried to use git commits to predict things like developer experience, coding time, and end-to-end implementation time, but instead of measuring these labels in the real world, he asked 10 "expert raters" to GUESS the answers. Not only that, 7 of these raters are actually managers/execs, which means they may not be touching code for quite a while now. The sample size is also really small, with only 70 commits, and the raters don't even agree on metrics like the developer's experience, code maintainability, and good structure. Still, the author took those responses and used them to train a Random Forest model to predict similar responses for new commits -- the only problem is that Random Forest models CANNOT READ CODE, so it's IMPOSSIBLE for this model to have ANY understanding of code quality
7400
-
7300
-
7200
-
7100
-
6900
-
6900
-
6500
-
6400
-
6100
-
6100
-
6000
-
5800
-
5800
-
5800
-
5500
-
5200
-
5100
-
4800
-
4800
-
4400
-
4300
-
4200
-
4200
-
4000
-
3900
-
3900
-
3700
-
3700
-
3700
-
3400
-
3400
-
3300
-
3300
-
3200
-
3200
-
3200