Is there a roadmap of capabilities this library currently or intends to provide?

[aaelony]

A number of Python libraries (e.g. a few named at https://www.pywhy.org/ ) implement various capabilities that fall under the umbrella of causal methods.

Is there a document that outlines current and planned capabilities for deepcausality? Perhaps a matrix of capability by implementation status? Is it a goal to achieve parity with the python ecosystem?

[marvin-hansen]

Well,

the thing is, I am knee-deep into a large internal project that will integrate the deep causality crate later this year.
At that point, I will probably fix the lifeline issue and sketch out what needs to be added and resume active development of the crate. I am really bad at making time estimates, but at some point its going to happen.

To your point, there is an interesting development that
a) more projects pop up with causal methods although rarely over geometric structures since most stick to the algebraic roots introduced by the do-calculus
b) the latest LLM models got somewhat better at causal reasoning although by no way deterministic and at fairly absurd price tag.

With all that in mind, it is most likely that over time I expand the usage of causal state machines as I have a future use case in advanced control systems were simple state machines don't make the trick anymore. Specifically, a buddy of mine works on codifying legal structures in smart contracts stored on blockchain and I actually designed causal state machines for exactly this kind of problem were reasoning capabilities are requires without a priori knowledge of what you have to reason about. This solves the problem of converting legally binding agreements into a deterministically verifiable causal state machines a smart contract can verify. But as said, that is scheduled for later as I am still building infra code.

Another area is almost certainly real-time stream analytics. I already wrote a blog post last year on the project blog. Meanwhile, I am integrating the iggy messaging bus, which has way better performance and latency, so at some point things like outlier detection over real-time streams will be relevant.

On a conceptual level, it should be possible to make the ordering requirement of the generic traits in the crate optional and thus allow operating on non-euclidian spaces thus bridging causal reasoning to neuro-symbolic reasoning thus forming the foundation for truly cross foundational reasoning. IMHO, that still has never been done so if you are interested in publishing a paper, that would be the way to go. DARPA had an open grant for work on neuro-symbolic research, so if you are US citizen, you can apply for that if you fancy. I am not therefore I can't.

The other thing is, if you feel there is something missing or could be added, feel free to open a ticket and share your idea or use case. I cannot vouch for anything for the time being, but I am going to take a closer look.