HN Explorer

Show HN: PyNeuraLogic: Python Differentiable Logic Programs (github.com)

37 points by LukasZahradnik ↗ HN
PyNeuraLogic is a framework that lets you use Python to create differentiable logic programs.

The framework offers an elegant way to express and further extend GNNs, as well as go beyond graphs and tackle other complex (relational) scenarios.

Feel free to check the repository and give us some feedback here or on Github. Thank you.

4 comments

[ 3.6 ms ] story [ 18.5 ms ] thread
melony
Congrats on shipping! I cannot wait to try it out, how does this compare to Scallop?

https://scallop-lang.github.io/ https://news.ycombinator.com/item?id=31060265

LukasZahradnik
Thanks! If you need any help with your trying out or have some ideas/bug reports, let us know on GitHub.

Scallop seems to be similar to PyNeuraLogic in the sense it extends logic programs with numerical parameters. The difference here is that PyNeuraLogic does not necessarily assign a probabilistic interpretation to values and parameters (which Scallop does, from my understanding).

versteegen
Super interesting! I think the time for this idea has come. I'm also working on a similar logic programming system with fractional value assignments, in Python, though without end-to-end differentiability as a goal. I'm amused to see you renamed "atom" to "relation" just a few days ago; I've found settling on clear and exact terminology to be especially hard! I'll definitely study PyNeuraLogic.
LukasZahradnik
I am looking forward to your project. It sounds interesting, and hopefully, you will share it here.

I agree. Determining the best and exact naming is hard. To be clear, "atom" was renamed to "relation" a few months ago. However, the documentation still contained occasional references to "atom" in the text - which was fixed a few days ago, as you mentioned.

If you need any help, let us know on Github.