program-induction
program-induction copied to clipboard
lorepozo
A library for program induction and learning representations.
Right now, `TRS` is declared as: ``` rust pub struct TRS { pub(crate) lex: Lexicon, pub(crate) utrs: UntypedTRS, pub(crate) ctx: TypeContext, } ``` It would be nice to have a...
The TRS sampler for terms, rules, TRSs, etc. is currently not very good. The difficulty is that it rarely samples things of moderate length. It either favors very long terms...
It would be nice to create a version of `weighted_sample` that takes in log probabilities. This might require the uniform to exponential trick described [here](https://stackoverflow.com/questions/33738382/).
TRS rules are currently non-deterministic, but they aren't probabilistic. It would be nice to make them fully probabilistic and to fit the probabilities during learning.
In the same way that the crate currently has lambda calculus, TRSs, and PCFGs, it would be cool to explore learning in some other representations, including graph rewriting, where the...
In the same way that the crate currently has lambda calculus, TRSs, and PCFGs, it would be cool to explore learning in some other representations, including conditional rewriting, where each...
In the same way that the crate currently has lambda calculus, TRSs, and PCFGs, it would be cool to explore learning in some other representations, including higher-order rewriting.
In the same way that the crate currently has lambda calculus, TRSs, and PCFGs, it would be cool to explore learning in some other representations, including Parigot's (1992) lambda-mu calculus.
In the same way that the crate currently has lambda calculus, TRSs, and PCFGs, it would be cool to explore learning in some other representations, including Barry Jay's (2016) lambda-SF...
In the same way that the crate currently has lambda calculus, TRSs, and PCFGs, it would be cool to explore learning in some other representations, including Barry Jay's (2009) pattern...