<pmikkelsen>Hi #guile! I am having a hard time with open-input-output-pipe from (ice-9 popen) <pmikkelsen>I want to communicate with an interactive program, but after I read 1 line the program I interact with terminates <pmikkelsen>is this desired behavior? the program is waiting for input when i try to read from it. <amz3>I need to choose between the graph based database with traversal <amz3>and the lower level eav database with minikanren querying <amz3>that's an issue I don't how to solve <amz3>the minikanren code is smaller <amz3>I am not sure of the complexity of minikanren quering tho <amz3>One thing I could do is rewrite what I have done using the traversal querying language using minikanren <amz3>dmiles: what do you think? <amz3>dmiles: is minikanren or graph traversal better? <amz3>so maybe I need to benchmark that stuff <amz3>that's not only a performance issue tho <amz3>I need the most easy and powerful API <amz3>to get started I will benchmark both grf3 and tuplespace over conceptnet.io <dmiles>amz3: i'd like to hear the results of your benchmarks <amz3>dmiles: I did not do them <amz3>dmiles: re-do them I mean. I am sure there is an overhead while using the graph structure for representing some information like lists <dmiles>oh thats what you meants for GraphDB <dmiles>well god *i* wonder now if that is what a lot of peopel meant by it over the years.. wow <amz3>I mean for instance, when I build trigrams out of a word, it requires an edge per trigram whereas the eav model doesn't require the "edge"... <dmiles>my opinion is that kind of graph database is one of the weorse case datastrores :p <amz3>it a simple foreign key, but at the end of the day, I could also implement it directly on top of the key/value store <amz3>which layer/vocabulary is the good one to best express my ideas? <dmiles>when people have said graph database i secrelelty immagined a database composed of all hashmaps <amz3>dmiles: that's graphdb on top eav <amz3>that's basically what I used to do <dmiles>had a branch leading to another branch finally maybe leaves? <amz3>dmiles: i have another question, are you familiar with “wikification” ? <dmiles>hrm god i guess i just described a list <amz3>eav is basically a list of hashmaps <amz3>dmiles: how do you do it? <dmiles>oh i am publishing a wikification of the prolog database state <amz3>dmiles: in particular how do you wiki-fy things like "Bob Marley" or "Barack Obama" and stuff? <dmiles>i pull up all assertions concerning "Bob Marley" onto one page <dmiles>and display them creating what spiders called doorway page <dmiles>to make the ulitmate page about Bob Marley <dmiles>also i make small queies into the db as well <dmiles>and display isa(iBobMarley,mobArtist). as "He was an artist" <amz3>I asking about your NER algorithm <dmiles>(That is how I and Cyc wikifies) <amz3>wikification is a super problem of NER <dmiles>but i guess i just see it backwards <dmiles>i dewikify (convert the data i read off wikipedia to named entities) by stealing first its subjects <amz3>basically the input string is "Bob Marley plays the guitar" I want to output '(PV (SUBJ "Bob Marley") (V "plays") (OBJ "the guitar)) <dmiles>oh I call that POS tagging.. well i do a virterbi "best guess based on fit" <amz3>you link both "Bob" and "Marley" together into a single entity? <dmiles>Marly is capitalized and adjacient to Bob so that is good indicator and huamn hame <dmiles>(and "Bob" is part of human names) <amz3>That's what I thought but it's not robust enough for non-human-name entities <dmiles>but jsut most people only make toy version of HSPG <dmiles>i've done non ty versions.. and never hit a bittle spot <amz3>dmiles: is there wikipedia page for HSPG? <dmiles>i mena it can be made robost to the point of insanity (genious) <amz3>hmm that's what I was thinking about a toy version of HPSG <dmiles>i did a semi-probablistic version once.. i even went so far as to use n-grams to help vote <dmiles>but it started as toy and just kept gettign more and more complex/robust <dmiles>why they are so neat.. they work almost on day 1 <amz3>dmiles: is that feature structures in the HSPG wikipedia page ? <amz3>dmiles: another question, how do you store all the wikipedia in prolog? <dmiles>yes.. for example .. doing number agreement <dmiles> are the fish numerous? is the fish only one? <dmiles>"are" agrees with plural versions <amz3>ah yes, you virtiby parser to match the sentence <amz3>that's what LG or RelEx was using at some point <dmiles>the "sing" was "number [aghreement]" there <dmiles>anything to "almost" use prolog :) <dmiles>sometimes i will grab off the shelf POS tagger to get its opinion before i HPSG it myself <dmiles>in case it does something brilliant that i might not notice <dmiles>i then useit POS opinions to guess help guess (brute firce) my routines <dmiles>where on that page "[[txt, 'George'], [tag, staart, nnp, np]]" in e2c("George fell last year"). <dmiles>staart is a feature that was added basted on being the first word <dmiles>IA = interaction agenda .. that is awesoem to <dmiles>i wanted sometimes to see when the parser was being overly subjective in its ear <dmiles>(like goerge the football player) <dmiles> (CityNamedFn "George" RepublicOfSouthAfrica ) would have needed to be possibly known to be at war <dmiles>or perhaps in some peacefull sports contest <dmiles>i like to even use abstract things like that to be part of the parse phase <amz3>you are talking science to me :p <dmiles>George fell last year riding his mountain bike on pensilvainia avenue <dmiles>would be "Geoge Bush" as the subject <dmiles>RelEx would use turd polish caused by "pensilvainia avenue" <dmiles>vs my polish .. which uses the same semantic vote <dmiles> turd polish = robustification by making everything seem less brittle by weaking the system using "floats" <dmiles>i call it "turd polish" ussualy when proability is used to cover for the fact that 2 or less smeantic routines are used <dmiles>i call it "turd polish" ussualy when probability is used to cover up for the fact that 2 or less senantic routines are used to guess things.. i liek to use 10+ things to guess things <dmiles>turd polish might be covering for for the fact that "close on some webpage" is all that is used to learn the connection between "George Bush" and "Pensilvainia avenue" <dmiles>some NER systems are trained first on "close on some webpage" <amz3>tx again dmiles I must go afk, I will let you know if i get wikification (or whatever the name) working <amz3>dmiles: you are too high up in the air for me to understand that joke :) <amz3>dmiles: give me a few more month so that a climb at you :) <amz3>dmiles: what do you do when the computer is processing data for a long time? do you do multi tasking? <dmiles>yes and sometime compete to CPU to irc rant <dmiles>yes and sometime compete for CPU resources to irc rant * <dmiles>but also mostly i try coding in lue of the process being successful <taylan>hmm, a with-current-directory procedure would be neat <wingo>taylan: no support on the os level, sadly <taylan>wingo: is there a problem with a naive dynamic-wind implementation? <wingo>no, it's that "current working directory" is a kernel-level per-process concept afaiu <amz3>is there a procedure that allows to compute all partitions for a given list <amz3>for instance '(1 2 3) return '(((1) (2) (3)) ((1 2) (3)) ((1 2 3)) ((1) (2 3))) <amz3>my algorithm fuu is laking <amz3>subsets without overlap and taking into account the order of elements <stis>working with a C type checker for clambda <stis>C type inference is not too difficult <stis>this is for compiling a scheme representation of C to C <stis>The scheme representation is better than C though, it has sane macros and more scheamy with named lets and everythingis an expresison principle <stis>I want to add some extra features for type deduction and support for SCM types ***random-nickname is now known as random-nick
<stis>I kind of like rusts type system which is a swet spot between C and haskell <stis>I plan to incorporate more rust features as I develop it. I want to learn rust and this is my way of doing it <wingo>holy crap this unicode thread, what is the deal