On Wed, May 22, 2019 at 3:11 PM Vitaly Bogdanov <vsb...@gmail.com> wrote:
> Hi Ben, > > It seems analogous to what is done in Haskell to support backtracking >> and Prolog-type cut and so forth. There one does use an "external" >> system to manage historical records, but this external system is >> hidden behind the monadic interface, >> >> http://okmij.org/ftp/papers/LogicT.pdf >> >> http://hackage.haskell.org/package/logict >> >> This is all using continuations behind the scenes, and related to some >> discussions Linas and I had earlier about making Atomspace schema >> processing support continuations. If one is manipulating >> continuations rather than just functions, then maintaining and >> manipulating the external computation graph is "just" an efficient >> shorthand for computing what the continuations denote... >> > > Yes if all calculations was in Scheme then it would be possible, but some > value manipulation logic is implemented in bare C++ and this logic is > spreaded over atomspace code so we cannot capture it easily. It is the > reason why we need passing computation graph from GroundedPredicateNodes > through custom Atom wrappers. > This sounds like that old conversation coming back to life. Stop using GroundedPredicateNodes. Start using Values. That is what they are there for. That is what they were created for. FloatValue was created last summer, specifically for the neural-net/tensorflow project. Sorry Vitaly -- this happened before you got started with opencog, or just in the first few weeks after you joined -- there were long discussions on this mailing list where it was clear that GroundedPredicateNode would not work, that it provided the wrong interface into tensorflow. If you look at the git commit history, that is when the `update()` method first appeared on FloatValue - it was designed to solve exactly this problem. (I looked just now. Its pull req #1709 - I diligently avoided saying "tensorflow" anywhere, calling it "streaming" instead. It's from exactly a year ago.) -- Linas -- cassette tapes - analog TV - film cameras - you -- 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 opencog+unsubscr...@googlegroups.com. To post to this group, send email to opencog@googlegroups.com. Visit this group at https://groups.google.com/group/opencog. To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CAHrUA35RDLDXKcDjuNkAa-tb8LiXGjaHZNgf1N2BxYoHjSfBGQ%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.
