16
rezn0v
7y

A neural network that converts pseudocode into actual working code.

Project Type
Project idea
Summary

A neural network that converts pseudocode into actual working code.

Description
I don't know if it is possible or not but it would definitely be a good idea. As a python dev I'm limiting the starting point to python only. What do you think?
Tech Stack
Python
Current Team Size
1
Comments
  • 0
    I think worth to try java or c as well
  • 6
    why not define an easy to use pseudocode language and translate it
  • 8
    To work with something with so much regular, rigid structure via a neural net you'd need a crapton of training. You'd need to have layers that mimic the stages of a compiler. You'd need to give the thing a lot of working memory for parsing, etc. so you'd have to use an RNN (or, as an optimization, an LSTM), and a large one, which only increases the training complexity and time.

    This is what grammars were made for. But if you use a grammar you need to specify a definite syntax for your pseudocode. In which case you end up with Python :p
  • 5
    There is such a thing... It called python hahaha ok sounds like a cool idea.
  • 0
    if this is possible then developers don't need to learn languages they just put pseudocode their and converted into the code by neural n/w🤔🤔
  • 1
    Isn't Lua kind of that already?
  • 2
    So you want to write a compiler for pseudocode?
  • 1
    Seems kind of pointless as it defeats the point of it being pseudo code.
  • 1
  • 2
    This is basically just a transpiler from a pseudocode-like DSL to whatever other language you want it to end up going to.

    Loads of ways to approach this. 👍
  • 0
    @Brolls finally some optimism!
  • 0
    Do you mean converting it into an actual code or executing pseudocode to work normally?
  • 0
    I like it, if nothing else because it could be a cool tool for improving the correction of tasks in logic courses or doing quick testing (e.g. a recursive solution to some problem)
Add Comment