10
Mitiko
6y

My goal for 2019 is to write an efficient neural network library with streams and make it very very very customizable. I'll write in C#, cause fuck python script kids. My idea is to make a cool platform for new ideas

Comments
  • 0
    @sergiolarosa89 Like 1337 h4x0r script kiddies, but for python users who can't code but use the language for ML
  • 2
    Blame the retards ml wannabe, not us
  • 1
    @devTea I might have overreacted. I'm sorry
  • 2
    @Mitiko lol I’m joking 🙃
  • 1
    But... b... but I love python and neural networks👨‍🎤
  • 0
    @-ANGRY-CLIENT- I have no problem with people who use python. I dislike people who only "learn" python because of NN libraries. Python is not the fastest language. But if you're a CPP programmer you are hardcore enough to write your own NN.

    There are plenty of JS and Java libraries. That's why I'm building my customizable nn in C# (ML.NET is not cool for neural networks)
  • 2
    Like fann? (http://leenissen.dk/fann/wp/)
    Or OpneNN? (http://www.opennn.net/)

    What actually *new* do you think you can invent?
    Writing it in C# will only accomplish one thing: Making it extraordinarily slow.

    Sorry if I sound harsh. I just fear that you will spend many hours for a result that will be nowhere near where you dream it to be...
  • 1
    @Yamakuzure You're not too harsh, I've also considered it can be a bit slow.
    My idea is to allow people develop new algorithms, while still using the neurons and layers idea of the neural network. For example instead of doing ax+b (a is a weight and b is a bias) we can do mx^2 + nx + b (m and n are weights and b is a bias)

    I had that idea (x^2) half an year ago, so I had decided to built a generic NN library. My goal for 2019 is to do it the right way (I tried once but my tests, the code and everything was terrible)

    I don't need it to be extremely fast because people would be writing proofs of concept and then maybe write their own faster implementation.
Add Comment