Hi Mani, Make sure you git pull build and reinstall all prior dependencies, including cog-utils, atomspace, the unifier, and URE. There have been significant changes in the last few months.
I'm confused about the comment about cython. I get `cython3 --version` 0.29.32 and cython never got above 0.2xx never mind something close to 1.0 Perhaps you're using cython 0.30.8 ?? -- "should work", famous last words. The PLN build should have worked. The github circleci test uses ubuntu 24.04 and last time it ran, it passed: https://github.com/opencog/pln https://app.circleci.com/pipelines/github/opencog/pln Its unmaintained code but it's also not bit-rotting. It should work the same as it used to. There were multiple problems with PLN, the biggest perhaps was that it was slow. A nit-pick critique is that It should use RuleLink instead of the much heavier and slower BindLink. This is a historical design artifact, but it had a huge performance hit. A theoretical issue is that It was built on the assumption that expectation values propagate, and they don't. People who work on stochastic differential equations deal with this situation in one of two different ways: either they take a million samples, and analyze the resulting distribution, or they take a highly technical, theoretical approach of looking at filtrations (sequences of sigma algebras) and the cylinder sets and measures thereupon. The sample-a-lot solution would have been to use crisp-logic inference (e.g. with SAT solvers or ASP solvers or even prolog solvers) and try every true-false combo possible, weighting them according to the PLN formulas. The benefit of this is SAT/ASP are blindingly fast, so you can map out a million possibilities quickly. The so-called "probabilistic programming" people tried something like this, slicing the cake in a different way. The technical approach is harder, but you'd need portions of it anyway, just to get the right "sample a lot" answer. I have a vague inkling that it rubs elbows with some of the deep-learning approaches, but I get distracted by side-quests whenever I start to think seriously about such things. There are soooo many interesting things to work on. It's a bit like being in a candy shop: each theory looks more delicious than the last one. Anyway, my impression was that PLN in the literal sense wasn't that useful (for anything I do.). In the metaphorical sense, where collections of sensori-motor systems or agents self-organize near a critical point, interacting with one another by sending snippets of PLN rules (or snippets or prolog-rules, or snippets of predicate-logic-type interconnections with each other, ... I'm agnostic about which) yes I think this is where rules systems get interesting. Think "a-life" (artificial life) or perhaps Conways-game-of-life, but instead of being on a square grid, it's on a random grid, and instead of exchanging single bits, exchange rules or rule-sets. (and of course, the random network reconfigures, because some agents will want to communicate more than others.) ***This*** is where I think the fun stuff is. If you get the last paragraph, then read through Michael Levin's "Ingressing Minds" paper, take it as an inspiration (he's a biologist, not a compsci/math guy), and now apply it to collections of simple sensori-motor systems (simple systems that can perceive structures, and move through them. For example "perceive file systems" (I think that's more interesting than vision, sound, walls of text, because it disposes of the regularity, and goes for the throat of recursive structure.) and then move around through that space, interacting with it, as autonomous agents. (Again, instead of 3D minecraft, move through an abstract space of relational structure.)) Whatever. I think it's cool, but I find it very difficult to make forward progress. It gets abstract and complicated really fast. But still, this is where I've been poking, when it comes to "doing stuff with rule-based systems". -- linas On Wed, May 14, 2025 at 3:20 PM [email protected] <[email protected]> wrote: > > Hi, > > On ubuntu 24.04 (Python 3.12 , Cython 3.0.8) I had to hack FindCython.cmake > for it to accept the Cython version. Also loading the PLN module in guile > gives the following error: > > In procedure dlopen: file "/usr/local/lib/opencog/libpln-types.so", message > "/usr/local/lib/opencog/libpln-types.so: undefined symbol: > _ZN7opencog22IMPLICATION_SCOPE_LINKE" > > Does PLN being abandoned means this kind of errors gonna grow more overtime > and I should think of something else ? > > Thank you, > Mani > > -- > You received this message because you are subscribed to the Google Groups > "opencog" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To view this discussion visit > https://groups.google.com/d/msgid/opencog/5b90c7dd-16f7-44cb-887d-188e35cea02en%40googlegroups.com. -- Patrick: Are they laughing at us? Sponge Bob: No, Patrick, they are laughing next to us. -- You received this message because you are subscribed to the Google Groups "opencog" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion visit https://groups.google.com/d/msgid/opencog/CAHrUA35NiN3ctNnigg2gVfEta58hv%2BjRJ1uY3jj2peyviaAyow%40mail.gmail.com.
