State of ML libraries?

Learn
#1

Hi! So, part of my bachelor’s thesis is basically running a CNN over some images that can hold letters, numbers, and a couple of logic symbols. This being my first time ever handling an ML task, I wanted to go safe with pytorch.

The thing is, I’m having a couple of type issues with the library, and I don’t really understand much about what the functions should get as input, or what they return… So, I thought that maybe a more statically typed programming language might help, hopefully, my favorite one :).

I’m looking for library recommendations that:

  • Allows me to write a simple CNN at first (like LeNET) and then maybe jump into other architectures (AlexNet maybe?)
  • Can run on an AMD Radeon GPU
  • Runs on windows? (not mandatory, but desirable since my it’s my tutors OS)

Thanks in advanced :slight_smile:

2 Likes
#2

The classic recommendation would be hasktorch, but in my opinion it’s not massively beginner-friendly (but it is very interesting!)

But, as a complement to also playing around with pytorch, you might find it an interesting/useful :slight_smile:

5 Likes
#3

Thanks for the reply! after asking around a bit, most people (teachers, advisor, and some friends) recommended me sticking to pytorch in order to finish my thesis as fast as possible, and then give hasktorch a try. So I think I’ll do that c:

2 Likes
#4

Yes, finishing your thesis with whatever does the job fastest is most recommendable. Then I’m sure you’ll have lots of fun with the haskell ecosystem once you’re done :slight_smile:

1 Like