Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Ask HN: What's Prolog like in 2024?

news.ycombinator.com

Post reply on HN
Print view View original

131–140 of 289 posts

Re: Ask HN: What's Prolog like in 2024?

#131

Post by ted_dunning »

inkyoto wrote: post #121
GistNoesis wrote: post #76

The "magic" of Prolog is built upon two interesting concepts : Unification ( https://en.wikipedia.org/wiki/Unification_(computer_science)... ) and Backtracking ( https://en.wikipedia.org/wiki/Backtracking ). Often bad teachers only present the declarative aspect of the language. By virtue of being declarative, it allows to express inverse problems in a dangerously simple fashion, but doesn't provide any clue for a so…

A unique property of Prolog is that, given an answer, it can arrive at the original question (or, a set of questions – speaking more broadly). Or, using layman terms, a Prolog programme can be run backward.

To be precise, a small number of very small Prolog programs can be run backwards.

There are essentially no significant Prolog programs that are reversible with acceptable efficiency.

Re: Ask HN: What's Prolog like in 2024?

#132

Post by ted_dunning »

dkarl wrote: post #69
greenavocado wrote: post #42

Earlier quoted context omitted.

Prolog and Datalog example (they are identical in this case) % Facts parent(john, mary). parent(mary, ann). parent(mary, tom). % Rules ancestor(X, Y) :- parent(X, Y). ancestor(X, Z) :- parent(X, Y), ancestor(Y, Z). % Query ?- ancestor(john, X). The Prolog code looks identical to Datalog but the execution model is different. Prolog uses depth-first search and backtracking, which can lead to infinite loops if the rules…

> Prolog uses depth-first search and backtracking, which can lead to infinite loops if the rules are not carefully ordered Is this an issue in practice? Most languages can create programs with infinite loops, but it's easy to spot in code reviews. It's been over a decade since I encountered an infinite loop in production in the backend. Just wondering if the same is true for Prolog.

Infinite loops in Prolog can appear with very subtle changes in the use of code.

One of the core problems is related to the reversible nature of Prolog. Not only are some programs reversible and some are, practically speaking, not, there are many gradations on this.

The result is that programs that look equivalent and whose tests appear equivalent may exhibit non-termination in surprising ways. This is, in my experience, the rule rather than the exception with Prolog.

Re: Ask HN: What's Prolog like in 2024?

#133

Post by ted_dunning »

dmpk2k wrote: post #78
greenavocado wrote: post #42

Earlier quoted context omitted.

Prolog and Datalog example (they are identical in this case) % Facts parent(john, mary). parent(mary, ann). parent(mary, tom). % Rules ancestor(X, Y) :- parent(X, Y). ancestor(X, Z) :- parent(X, Y), ancestor(Y, Z). % Query ?- ancestor(john, X). The Prolog code looks identical to Datalog but the execution model is different. Prolog uses depth-first search and backtracking, which can lead to infinite loops if the rules…

How does the Datalog approach compare with RETE?

The big deal about Datalog is that it is equivalent to SQL-with-recursion. Thus, it can compile to database queries.

Re: Ask HN: What's Prolog like in 2024?

#134

Post by vector_spaces »

To piggyback onto OP to ask about something very loosely related: what about miniKanren? Are there any active projects or work being done here, either in academia or industry? Most of the ones listed on minikanren.org appear to be dead -- although I haven't gone through them all since last year

Re: Ask HN: What's Prolog like in 2024?

#135

Post by GistNoesis »

abeppu wrote: post #119
GistNoesis wrote: post #76

The "magic" of Prolog is built upon two interesting concepts : Unification ( https://en.wikipedia.org/wiki/Unification_(computer_science)... ) and Backtracking ( https://en.wikipedia.org/wiki/Backtracking ). Often bad teachers only present the declarative aspect of the language. By virtue of being declarative, it allows to express inverse problems in a dangerously simple fashion, but doesn't provide any clue for a so…

> The continuous equivalent in a sort of unification is Rao-Blackwellisation (done automagically by deep-learning from its training experience) which allows to pick the right associations efficiently kind of the same way that a "most general unification algorithm" allows to pick the right variable to unify the terms. I don't know how to reconcile this statement about deep learning with my understanding of Rao-Blackwe…

Once you view prolog goal reaching as a game. You can apply Reinforcement Learning methodologies. The goal is writing a valid proof, aka a sequence of picking valid rules and variables assignment.

Value being estimated : The expected discounted reward of reaching the goal. The shorted the proof the better.

The sufficient statistic : The embedding representation of the current solving state (the inner state of your LLM (or any other model) that you use to make your choices). You make sure it's sufficient by being able to regenerate the state from the representation (using an auto-encoder or vae does the trick). You build this statistic across various instances of problems. This tells you what is a judicious choice of variable based on experience. Similar problems yield similar choices.

The crude estimator : All choices of have the same value therefore a random choice, The improved estimator : The choice value is conditioned on the current embedding representation of the state using a Neural Network.

You can apply Rao-Blackwell once again, based by also conditioning one-step look-ahead. (Or at the limit applying it infinitely many times by solving the bellman equation.)

(You can alternatively view each update step of your model, as an application of Rao-Blackwell theorem on your previous estimator. You have to make sure though that there is no mode collapse.)

You don't have to do it explicitly, it happens under the hood by your choice of modelisation in how to pick the decision.

Re: Ask HN: What's Prolog like in 2024?

#136

Post by everforward »

jodrellblank wrote: post #116

What is it like? 50 years of historic cruft. Questionable whether there are more trip hazards than usefulness for ordinary coding. A fractured community which feels like there are more Prolog systems than Prolog code. Learning Prolog is less "how do I do things in Prolog" and more "how do I contort my things to avoid tripping over Prolog?". A few dedicated clever people and idealists and dreamers talking about ontolo…

> A few dedicated clever people and idealists and dreamers talking about ontologies and building things I don't understand

I was briefly deeply interested in ontologies via OWL and I suspect Prolog has the same issues that I think plague ontologies in general.

They are a fantastic tool for a system complex enough to be nearly useless. Modelling an ontology for a reasonably complex domain is unreasonably difficult. Not because the tools are bad, but because trying to define concrete boundaries around abstract ideas is hard.

What is a camera? A naive attempt would say an item that takes pictures, but that would include X-rays. Are deep-space radio telescopes cameras? Trying to fix those issues then causes second order issues; you can say it’s something that takes images from the visible light spectrum, but then night vision cameras aren’t cameras anymore.

The reasoning systems work well, they just don’t solve the hard part of designing the model.

Re: Ask HN: What's Prolog like in 2024?

#138

Post by riku_iki »

everforward wrote: post #136
jodrellblank wrote: post #116

What is it like? 50 years of historic cruft. Questionable whether there are more trip hazards than usefulness for ordinary coding. A fractured community which feels like there are more Prolog systems than Prolog code. Learning Prolog is less "how do I do things in Prolog" and more "how do I contort my things to avoid tripping over Prolog?". A few dedicated clever people and idealists and dreamers talking about ontolo…

> A few dedicated clever people and idealists and dreamers talking about ontologies and building things I don't understand I was briefly deeply interested in ontologies via OWL and I suspect Prolog has the same issues that I think plague ontologies in general. They are a fantastic tool for a system complex enough to be nearly useless. Modelling an ontology for a reasonably complex domain is unreasonably difficult. No…

your camera example demonstrates that human knowledge is loosely structured and formalized in general, so you can't create strict onthology. One way to work around is to assign some confidence score on statements, so you will have something like that Nikon device is likely camera, and x-ray machine is unlikely camera based on current world model.

Re: Ask HN: What's Prolog like in 2024?

#139

Post by btbuildem »

rramadass wrote: post #50

Though i only know Prolog cursorily it is in my todo list of languages to study. I think it has great value in that it teaches you a different paradigm for programming. You might also want to look at Erlang which is used in the Industry and would be helpful for your future. Joe Armstrong was originally inspired by Prolog and he conceived Erlang as Prolog-Ideas+Functional/Procedural+Concurrency+Fault-Tolerance. Hence…

Ha! That explains a lot. I've started looking into Prolog recently, and there were some... familiar echoes in there, reminiscent of Erlang.

But of course, the submarine is like a cigar, not cigar like a submarine.

Re: Ask HN: What's Prolog like in 2024?

#140

Post by btbuildem »

sproutini wrote: post #46

The problem with Prolog is that it's based on unification, and small unification engines can be expressed in a few lines in any functional programming language. That narrows down the already small niche where one would choose Prolog by probably a few orders.

For some, how a language is implemented seems to be the paramount thing.

For many, how the language faces the user, how its paradigms fit the problems at hand and the user's mode of seeing the world, that is more important.

These days, with the terabytes the petaflops and the megajoules, it might be even less relevant how the gears are turning inside the black box.

Post reply on HN