News for hackers. Stuff that matters.
Re:MLJ.jl: A Julia package for composable machine learning by Tarrosion on Monday April 12, 2021 @10:59AM (#26780123)
I want to love MLJ and I _do_ love Julia, but holy wow is it hard to learn. I think three times now I've had a small dataset I wanted to do something simple like linear regression on, thought it'd be a good opportunity to learn MLJ, and ended up giving up when I was knee deep in inscrutable errors about scientific types and machines and unsupported models.

If anyone has a good introduction to recommend which is clearer than the official docs, I'd definitely appreciate it.

Re:MLJ.jl: A Julia package for composable machine learning by gugagore on Monday April 12, 2021 @11:19AM (#26780352)
Sorry, I don't have a good introduction to recommend, I just wanted to include a link if anyone is intrigued by "scientific types", which refers specifically to: https://github.com/alan-turing-institute/ScientificTypes.jl
Re:MLJ.jl: A Julia package for composable machine learning by indeedmug on Monday April 12, 2021 @06:50PM (#26786159)
The MLJ library has some example notebooks in the github repo somewhere. I used those notebooks to figure out the API of MLJ because there are unstated things in the documentation. But I agree, it's brutal to learn Julia because the documentation is so lacking for beginners and type errors can be very challenging to understand.
Re:MLJ.jl: A Julia package for composable machine learning by Tarrosion on Monday April 12, 2021 @10:43PM (#26787601)
FWIW I've found learning Julia (and using it daily) mostly overwhelmingly pleasant, and I perceive MLJ as an unfortunate outlier. Sounds like you had a different experience.
Re:MLJ.jl: A Julia package for composable machine learning by adgjlsfhk1 on Tuesday April 13, 2021 @07:36PM (#26801008)
If you haven't tried out Julia recently, you might want to. 1.6 makes some pretty big improvements in stacktrace printing.