FRDCSA | internal codebases | CRITIC

[Project image]

Architecture Diagram: GIF
Code: GitHub

Jump to: Project Description | Parent Description | Capabilities

Project Description

Here is an incomplete list of some things that critic will be used to browse. CPAN apt-archives freshmeat sourceforge systems (radar) plans (pse) recipes (gourmet) sources.list (apt-sources) In addition to things like these, critic should be capable of using collaborative filtering to make recommendations to users. This functionality could be used for instance to create models of people's interest in music and videos for posi meetings.


  • Finally create critic using kbfs and NLU.
  • KBS, query ("critic-unilang-classification" ?X ?Y)
  • KBS, work query ("critic-unilang-classification" ?X "goal")
  • KBS, query-cyclike ("critic-unilang-class" ?X)
  • KBS, query-cyclike ("critic-unilang-class" ?X ?Y)
  • KBS, query ("critic-unilang-class" ?X ?Y)
  • KBS, default query ("critic-unilang-classification" ?X "donation-request")
  • KBS, default assert ("critic-unilang-class" "donation-request")
  • KBS, default query ("critic-unilang-class" ?X)
  • critic many entries in order to determine their releasability through classify.
    ("depends" "100275" "100272")
    ("depends" "100277" "100275")
  • Get critic critique functinoality working.
    ("depends" "100275" "100272")
  • For all programs, we can look at abstractions of their functionality, to compute a better way of writing the program in the first place. In other words, for every program P, there exists a system S which executes P in a more intelligent, versatile way. I know that isn't formal, but here is an example. The script that rates the apartments, well, hand writing that is silly because, what if you want to rate something else, you see. The notion of a system that helps you rate things is important. critic ideally subsumes that...
  • Sorcerer should interface with critic or Review.
  • Gut critic so that it can't be used for porn.
  • Write TK script for critic to expedite classification of our entries.
  • Use picform Contexts for unilang classification, study assignment, etc. Eventually, develop Context classifier as part of critic?
  • critic and Context have lots in common?
  • In order to use critic in Emacs on Sorcerer data, simply send the list to unilang-emacs-client via some new function that converts it into an alist and browses it with some dired like editor.
  • Use the critic system on this score stuff as well, in order to come up with multiple taxonomies.
  • apply critic to study
  • (We then get critic::Classifier to use SQL for its entries source)
  • Setup collaborative filtering interface for different books (software, etc, critic.)
  • Write a simple perltk user interface to get used to it for critic.
  • critic should be able to take ratings and other constraints and properties set on objects and backwards infer their value.
  • critic can use the new KBS::Store module.
  • This new system is the basis for critic, kbfs (now we have the KB for it), gourmet's ontology, etc, etc, etc.
  • The critic system is almost complete using freekbs, and newly written code currently under score.
  • We will use a classifier to classify writings into topics automatically using the code I just wrote for critic.
  • This thing I've made for Sorcerer sure looks useful, it is being used in Sorcerer, broker, should be used in job-search, and could be used in busroute, clear, corpus, critic, cso, digilib, and verber.
  • boss QC Use best practices - verified in perl code using Perl::critic, etc.
  • Send email to the guy who gave the Banshee talk at Sat Sep 20 15:54:06 CDT 2008 information about a more expressive way to rate music (than 1 - 5), you know the whole critic comparison langauge...
  • Add key bindings for critic mode stuff when not in critic
  • Implement a 'more goals like this' feature for critic.
    ("pse-has-property" "66412" "important")
  • Fix state problems with critic Knowledge View mode.
  • Have an easy way to edit stored knowledge, maybe with the critic knowledge editor?
  • Force myself to classify 1000 entries as a start, finish the critic classifier subsystem.
    ("due-date-for-entry" "60341" "Important")
  • critic needs to model what the user has already seen.
  • critic needs to be able to show information about the entries, not just the entries themselves.
    ("completed" "60318")
  • Implement critic now that freekbs is working
  • Use critic each day to manually review entries.
  • I think the most important thing for me to get working right now is the class browser for critic.
  • critic::Classifier::UI should clearly take into account that certain classes subsume others.
  • critic::Classifier::UI should have interfaces for partitioning, verifying groups.
  • Look into writing that type browser in tcl for critic.
  • However now I'm too tired to do this. So I need to do that and other related things, like build a better critic::Classifier type system for corpus, tomorrow. But I must also eat tomorrow.
  • I should set up RSR to start recording so I can begin to get some feedback on my situation. A great solution is to use critic::Classifier on RSR::Event s.
  • RSR can use the critic::Classifier system to classify its events into habits.
  • critic may be applied to RSS feeds.
  • critic should generate assertions that could be added to Cyc, for instance.
  • Use opencyc to maintains lists of objects and critic should of course interface with opencyc.
  • Use critic to relate different Perl modules to different items in our system.
  • Probably the most important thing to do is to always make concessions. For instance, no pornography. I suppose, obviously, one would destroy the critic system if this were this case. I also will have to institute punishments for violating the rules. This means that, frequently enough, I must take the entire survey. I don't know why I feel good right now. I think its mental.
  • I have written that one system and it is /var/lib/myfrdcsa/codebases/internal/critic/learning-classifier.pl: perl script text executable
  • Add refresh delays to critic.
  • I ponder deleting the etnire critic system.
  • There should be a module for forcing users to review systems, i.e. forcing them to use critic on Sorcerer.
    ("pse-has-property" "1273" "habitual")
  • As radar finds systems that interest the user, it presents these systems as though in a flight simulator - perhaps using critic, and feeds documents that should be read to clear
  • A major problem is that with systems like critic and clear we still wind up deleting the information because they either lose information or are not sensitive enough, respectively.
  • critic repeats things too much before exploring more.
  • (critic = WEKA more or less)
  • need to find critic/ semi-supervised/ collaborative filtered stuff soono
  • Can even use information from critic.
  • Could use a temporal tagger, as well as critic to rate the experiences.
  • critic has some of the beginnings of a source recommendation system, and a reputation management system. social knowledge is critical to advancing our cause.
  • is it even possible to critic AI as though it were one idea?
  • basically critic is like a knowledge base editor with some machine learning features to suggest knowledge by analogy that may be useful, and basically to help the user alter the state of relations pertaining to a particular system.
  • got the beginning of disciple and improve critic dramatically. critic now seems to be geniunely working, although we should add the ability to infer from time spent looking at an item what the user's interest is, put an interval of expected rating, distinguish between the internal rating and the users rating, maybe call it utility, and add time-dependent features to the analysis. Also look at carbonell's javelin algorithm as a metric.
  • My work with the critic system has made me extremely bitter. I keep trying to fix nonsense bugs, but the system is too unwieldy. I'll think its something and modulate this back and forth, trying all possibilities, until it becomes obvious that the problem is somewhere else. This is a lesson, mainly, that we will have to break tasks down into more manageable chucks and verify them. Divide and conquer. The ideal way to do this therefore involves listing tasks, decomposing them, and verifying the completion of each subtask manually. Inductively, so that when the system is run, we methodically check all possibilities until we have eliminated them. I.e. we must be able to prove correctness of plans, etc. But in order to even implement this ability, we will have to apply these rules manually until we have some systems to do this. This means interactively building up a system from small, but verified components.
  • some criticism of the critic system. It is working better now, but it still has problems with its ratings, as in they are not accurate. must fix problems with negative ratings, and no ratings. could add the ability to use previous durations to calculate new expected duration and set a timer watcher.
  • should add a feature to critic where it looks at sources deemed similar to highly rated ones first.
  • critic of planner: looks nice, but as usual, these systems are based less and less on data and more and more on approximations. I think our system is necessary to provide the data driven aspects of planning.
  • still need a better system for interactively analyzing relationships (critic)
  • I finished and tested the basic critic architecture, (that's why my typing is so bad right now.) It turns out to be ok, but really slow.
  • radar search results should be scored and used by critic, etc.
  • perhaps have critic have an expanded as well as contracted view of the metadata file.
  • critic.el can communicate with critic.pm through unilang.
  • use this as one rating scale that can be applied to CPAN, for instance, and read through critic. So obviously critic should support various rating scales, etc.
  • critic could possibly even use unilang to communicate, say, task dependencies to pse.
  • it looks as though the early dialog management system will rely somewhat on critic.
  • use the metadata for the cache as input to critic, it was funny, cause I was thinking how arbitrary it was to simply save the cache metadata as an XML file, and that the cache metadata file ought to do something, and then, quite naturally, we have the concept of feeding it to critic.
  • use critic to enable classification of data
  • use critic to rate words as being important, and use this information, as well as others, to help digest find relevant application data on the web.
  • use critic, or whatever, to match a given entry to its normalized form.
  • use heterogenous, semistructured extraction tools for critic, create wrappers for various sources, like rdf dump, xml file, database output, etc.
  • the notion of explore versus exploit of critic also applies to things like ITS.
  • if desired critic should be able to make detailed comparisons between n items.
  • ai::categorizer is useful to critic
  • relavent to critic http://www.isi.edu/~blythe/evaluation-ontology/
  • expand the radar Criteria system to interface somehow with critic, I would imagine they are somehow related.
  • critic will be dangerous and difficult to write. Maybe make it into an emacs system, or write an abstraction. Is critic really necessary for pse? I don't think so. Really, just move the frdcsa-perl shit into radar, install those pse mods, read their manuals and be done with it. Then you can hanker down on pse again!
  • get unilang search capabilities going. use critic on unilang!!!
  • or choose a program that does blah blah blah. critic can thus be used by radar to judge the value of potential software systems it should package.
  • in that way the author of some program simply calls critic on tasks like "choose a recipe that the user wants", "choose a healthy recipe", etc.
  • The gist is that critic can also be used to critic plans
  • I am starting to imagine some real interplay between critic and pse
  • dselect should be a related work system to critic
  • the notion of critic applies to clairvoyance and its notion of tracking what a user has seen, as well as kbfs, and also relates to the general properties of exploration, for instance, in searching for documentation about a particular capability. what is the name of the capabilities management system? cando?
  • here are some dependencies: pse unilang ?, pse myfrdcsa::Conf ?, pse critic?
  • does pse depend on critic?
  • is critic the most important?
  • use new techniques to finish the critic system
  • critic relies on perl programming to finish
  • need to finish critic program to do this
  • dpkg-iasearch could be used as a background related-work system for critic

This page is part of the FWeb package.
Last updated Sat Oct 26 16:51:05 EDT 2019 .