Case Mapper - Qualitative reasoning group
Contents
1. 8 2 2013 parentheses If your file has mismatched parentheses it will not load in Case Mapper We recommend Emacs a freely available text editor How to Define new predicates and concepts To define a new predicate you must specify several things via facts in meld files 1 Is it a relation or a function If it is a relation then you should include the statement isa lt new gt Relation in your file If it is a logical function then include isa lt new gt Function Denotational The collections Relation and Function Denotational are used by the KB SME interface to determine whether something is a relation or a function The minimum additional information that SME needs is its arity number of arguments For fixed arity relations or functions arity lt new gt lt integer gt where lt integer gt is the arity suffices For variable arity relations state isa lt new gt VariableArityRelation and for variable arity functions include isa lt new gt VariableArityFunction Is it an attribute Attributes are implemented as collections i e concepts so you should include the statement isa lt new gt Collection in your file You should also incorporate your new concept into the existing ontology to help keep things conceptually clear You may find out in the process of doing so that a reasonable concept already exists and hence you don t have to define a redundant one To do this find a relevant con2. 0005 e Local score for identical function The local score for a match hypothesis which has identical functions This is typically smaller because functions representing dimensions can easily match non identically if they support a larger matching structure The default value is 0 0002 e Trickle down multiplier Systematicity is implemented via a trickle down algorithm where the score of a match hypothesis cascades down to the match hypotheses of its corresponding arguments The default value is 8 0 e Maximum local score for a MH Provides the ceiling for match hypothesis scores The default value is 1 e Minimal ascension multiplier Indicates the penalty to be imposed if the alignment is due to minimal ascension The default is that only 8096 of the score will be considered Experimental Parameters e Functor trickle down Allows trickle down to match hypotheses connecting functors e Require overlap for merging An experimental new version of greedy merge which requires kernels to overlap Normally all that is required is that they are not structurally inconsistent e Use global candidate inference support Use global mapping information when calculating candidate inference support scores Normally scores are computed for the correspondences on which the inference rests locally e Use less greedy greedy merge Use an experimental form of greedy merge allowing addition of consistent kernels from other mappings 26
3. MAC FAC Browsing Reminding entitySupportedinferencesNotAllowed reminding base 13 SyslitCaseLibrary TheSet sme r Probe base 13 Case Library SyslitCaseLibrary With Selected Case ta 5 ta 7 ta 8 ta 10 ta 12 SME Compare to Probe CV Compare to Probe Run MAC FAC MAC Results The following results are for the MAC stage only Case Value Is 13 0 94152904 ma 13 0 88431877 The following cases MAC FAC Results are FAC analogy output given the above MAC results Case Matcher Is 13 Browse SME Figure 11 Viewing MAC FAC results and click SME Compare to Probe This will open a SME mapping browser discussed above between the probe and the selected case b Select a case from the Case Library list and click CV Compare to Probe This will display the content vectors of the probe and the selected case as well as the dot product of the two vectors This dot product is used in the MAC stage of MAC FAC c Click Run MAC FAC to run MAC FAC again and see the results 5 The FAC stage of MAC FAC is achieved via SME comparisons between the probe and the FAC Constraints KBCaseFn base 13 SyslitCaseLibrary 5 Available constraints identical functions Current constraints The Structure Mapping Engine can constrain its mappings in a number of ways These controls allow you to manipulate a subset of these const
4. SME for performing analogical comparison e Use MAC FAC for performing similarity based retrieval e Use KQML server for easy programmatic integration with external components For information about the algorithms behind these cognitive models and related experiments please see the following references 2 3 4 These can be found on the QRG web site at www qrg northwestern edu In learning Case Mapper we suggest that you work through the examples in the order provided in the manual About Knowledge Bases and Cases Case Mapper is always used with a knowledge base KB The KB stores knowledge including facts and cases Cases are simply collections of facts that can be treated as a unit Inside the KB cases are implemented via Cyc style microtheories Sometimes browsing tools will refer to microtheories and that s what they mean You can just think of them as cases Some facts in the KB specify properties of the vocabulary of predicates it uses including their arity number of arguments whether they are a relation or a logical function and what types of arguments they can take SME and MAC FAC require this information for any predicate used in cases It is important to understand that there is a translation process that occurs when facts are added to an SME description Recall that in structure mapping theory there is a sharp distinction between attributes which are unary predicates on entities and relations which have two or more argu
5. flag defaults to false but in learning systems built on top of SME it is often set to true e Allow out of mapping contributions This determines how the final mapping score is computed The default is set to true indicating that the mapping score is the sum of all MHs in the mapping e Block most out of mapping contributions The default is false When set to true checked this limits the initial MH score trickle down to mostly structurally consistent parents 25 8 2 2013 e Enforce 1 1 minimal ascension The default is true This allows functor correspondences to be created for relations matched via minimal ascension This in turn means that non identical predicate matches must be identical across the entire mapping e Compute reverse inferences This allows SME to compute candidate inferences from the target to the base as well as from the base to the target This is very useful in computing alignable differences Numeric Parameters e Maximum Gmaps The maximum number of mappings to compute The default value is 3 e Greedy cutoff value If there is a steep drop off in score from the best mapping lower scoring mappings are dropped This parameter sets the threshold for the cutoff i e subsequent mappings must be within this percentage of the best mapping The default value is 0 8 e Local score for identical functors The local score for a match hypothesis between expressions which have identical functors The default value is 0
6. need not retrieve facts before forgetting them we did this purely for demonstrative purposes 8 Verify that the fact was properly deleted from the KB by refreshing your KB browser window 20 8 2 2013 Next we discuss how to run SME using KQML ask messages 1 If you are starting a new Case Mapper or Lisp IDE session complete steps 1 3 above 2 In your IDE evaluate send test msg ask query matchBetween ExplicitCaseFn solar system ExplicitCaseFn rutherford atom TheSet match context cm test context The predicate matchBetween invokes SME and the predicate TheSet indicates that its arguments should be viewed as a set Here since there are no constraints TheSet is indicating the empty set The context cm test context indicates that the mapping will be performed with respect to cm test context This allows the user to customize which predicates should be considered for analogy using analogy control predicates 3 Case Mapper s response message should include the content ist Information cm test context matchBetween ExplicitCaseFn solar system ExplicitCaseFn rutherford atom TheSet MatcherFn 7 0 Note your MatcherFn 7 0 term may differ in numerical arguments this is only an internal identifier token that serves as a handle for getting further information about this SME operation while it is still resident in Case Mapper s working memory 4 Using the MatcherFn term you r
7. the expression correspondences link while browsing apto Detalle ExcfaaHon a mapping These expression correspondences are what Figure 7 Browsing a SME mapping suggest the entity correspondences Moreover since score trickles down to the arguments of a correspondence the expression correspondences provide support for the entity correspondences such that those entities which participate in large relational structures get higher scores How to Constrain SME By default SME will produce the best structural mapping or up to three if the others are very close for the given base and target However sometimes the context for analogical reasoning provides more constraints that need to be respected In understanding an explanatory analogy for example the context provides explicit constraints that some entities must match e g An atom is like the solar system means that atom and solar system must correspond In analyzing a complex analogy several 11 8 2 2013 iterations of mapping can be required e g 7 so SME must be told to avoid certain correspondences to force it to look for alternatives Case Mapper provides an interface for adding and removing such constraints To illustrate let us consider a variation of the mapping we just used Fetch cases base 10 and ta 10 from the SME Classics KB Select KBCaseFn base 10 as the SME base and select KBCaseFn ta 10 as the SME target The SME frame of your window will
8. you choose unblock since otherwise the program will not be able to communicate with the browser If you get a dialog box that says unrecoverable error when opening a KB it means that either there isn t a KB at the location you selected or it is corrupted Delete whatever files are at that location and restore your KB from the backup you made If you are browsing and you get something that says Lisp error instead of what you expected then you hit a time out in the browsing system Either the operation is just too big for the browsing system to handle e g showing all isa statements in the OpenCyc derived KB or your machine was spending a lot of time elsewhere In the latter case try again When you get an unrecoverable error dialog if you click on debug instead of ok the program will in some circumstances keep going just fine However you should save your KB soon and be sure to have a backup handy just in case If you have difficulty starting or connecting to the Case Mapper KOML server ensure that your Windows firewall is disabled at least for the port you have specified as your KOML server port Testing CaseMapper You can test the basic functions of CaseMapper using Help gt Test CaseMapper menu option This operation requires you to have smoke kb as the current KB So close any other KB and open smoke kb included with the installation A file selection dialog will pop up This lets you specify the name and
9. 8 2 2013 Case Mapper User Manual August 2013 Beta 3 Analogy Conference Edition Kenneth D Forbus Scott Friedman Subu Kandaswamy Andrew Lovett and Jeffrey Usher Qualitative Reasoning Group Northwestern University 8 2 2013 Contents VEO M 4 About Knowledge Bases and Cases cccssssssccececessessesecececesseeaaeseeeescessesaaaeeeeeessussesaeaeseeeessseseaaaeseeeeesesegs 4 How to Open a Knowledge Base etie aie ha piena se te ea idad cra ee na Dus ne aUa de aa via dee aa RR e aa Ee DES 5 How to Back up a knowledge base esses eene entren nnns seen rn sientes seen nnns 6 How to Close Case Mapper 2 oic baee dao etece top aeuo erae n oSe e aeo tst e aeneo caavtevesssaeedaatinde Seatac cion e canes 6 How to Browse a Knowledge Base esses eee enne entente en nnns sten nens seen nennen nens 6 HOW tO Fetch a Case ince Lm 8 Using SME from within Case Mapper ccessscccccecsssesenececececeesesseaeeececseessesaeaeeesecsssesaaaeaeeeeeeessesesaeaeeeeeens 10 Howto RUM IS ME dai 10 How to Browse SME Mappin ccccccceceeseeenneaeceeeceeeeeeaaeaeceeeeeseeeaaeaeeeeeeeeeeaaaeaececeeeeeeeaaeaeeeeeeeeneaaea 10 How to Coristrait SME soccer ret Er ERE OORE TONA Di IRE Tar Xo Ee DEREN ERE FERRE 11 Using MAC FAC from within Case Mapper cccscccssssecsscceescecessseessececssecessee
10. ML packet that sends the content ask query isa x Dog to Case Mapper tell sender localhost receiver localhost in reply to reply with MSG 2 language ontology content ask query isa x Dog Note that all of the keyword fields except for content are specified by character strings and the packet must contain quote characters around these parameters Conveniently you can use the above packet structure and substitute only the parameters for reply with and content to programmatically access Case Mapper The Case Mapper server accepts these packets in string format and its response packets are in the same format Note that the Case Mapper server socket connection is not persistent with its clients so it disconnects each time a packet is processed after a response packet is sent back to the sender In the following subsections we describe each interaction and packet Note that only the content field changes for each of the interactions How to Use KQML to Ask The above packet example with content ask query isa x Dog performs an ask operation on the Case Mapper server with the query Consequently Case Mapper will use its working memory and its KB to find all possible answers to the asked query in this case it will bind the variable x to all known instances of the collection Dog The response will look something like this tell sender localhost 18 8 2 2013 receiver lo
11. __ Run sme 3 Fetch the case ta 10 Target Case KBCaseFn ta 10 c Set Constraints 4 Select KBCaseFn base 10 from the SME Base Case dropdown list and select Figure 5 SME is ready to run KBCaseFn ta 10 from the SME Target Case dropdown list The SME frame of your window will resemble Figure 5 5 Click the Run SME button If no message appears notifying you of a problem SME has successfully computed analogical mappings between the base and target cases and the mappings are ready to be browsed If the web browser does not open automatically press the Run SME button again How to Browse SME mappings The SME browser opens automatically after running SME The browser interface provides an interactive report on the mappings that SME generated from the two representations Browsing SME does not affect the mappings that were generated this is just a manner of inspecting various details of SME s output aes We ll start with the initial page as shown in Figure 6 M deis aia Target KBCaseFn ta 10 which results from following the instructions in the last Mapping Score MHs Cls secti on Mapping 68 10 3780 84 3 16 entity correspondences The SME number 2 in Figure 6 and the Mapping ss number 68 in Figure 6 will likely differ from your 34 functor correspondences results these are identifiers used to differentiate various Mapping Parameters SME instances and mappings i
12. adding the fact arity relationInstanceExists 3 Assertion in microtheory Bob2 Problem Could not find a predicate definition for in microtheory Assertion relationExistsExists loves Person Cat Problem Could not find an arity for predicate relationExistsExists Try adding the fact arity relationExistsExists 3 Assertion relationInstanceExists anatomicalParts GenericInstanceFn Cat Mouth Problem Could not find an arity for predicate relationinstanceExists Try adding the fact arity relationinstanceExists 3 Note These suggestions are suggestions to make your representations syntactically sound and feasible for use with the Structure Mapping Engine You may still have semantic problems in your representation Figure 14 Sample Report from check meld You can test the syntax of your case file by choosing File gt Check meld file from the File menu This will produce a report page in your browser as shown Figure 14 For our example file in Figure 13 we have not defined the predicates relationExistsExists and relationInstanceExists KQML Server Case Mapper comes with a built in server module so that other applications and cognitive models can access SME MAC FAC and knowledge base programmatically Also included is an example client application in the file test client Isp This is a Common Lisp implementation of a stand alone program that sends and receives KQML Knowledge Query Manipulation Language packets to the Case Mapper
13. caeeeesaeeeesseceseeesateeeaeeeeaeeesaes 13 How to Select a Case Library iia 13 How to Run MAC FAC Quo ccccccccsecccccesssccceesssscueussceceeussseceeusesecuaususecueususeeuaususecueususecuausesecuauseseeseuseseeneneees 13 Extending knowledge Dase itte Co adan ina 15 How to Write a case file iet eret eee b ee eee e Lev ere En hoe beth eeu dete punte epu Eva Rada 15 How to Define new predicates and concepts ssssssssssssseeee esee eene nennen ener eterna 16 How to Load a case file into your knowledge base ooccconoccccconoccncnanonnnonananonocanonnncnanononcnanonnnccanannnccnnns 17 How to Test a case file eee ee epe ia 17 KOMIL S IgM tds 17 How to Start amp Stop the KOML Server c occcoooccccconoccnccononcnonanononcnonononcnanonnnccnnn nn nc ennt esent ne enne nnne ena 18 Basic KOML amp Packet Structure esses eene enne en nen nennen cnn cnn anna cnn 18 How to Use KOMLEL tO Ask inei ee ptr hori eere HERR Re eae UNSER ER Ea aane Ea ER Re eet d e ER Ren 18 How to Use KOML tO Store eere erret ea erre stet aeo rera te esee erp she otto kae o nsa cine 19 How to Use KOML to Retrieve essessesseseeeeeeeeneneneneneene nnne enne een nnne esent rennen enses 19 How to Use KQML to Forget eessssesssseseesesee EEE nennen nennen tenete seen esent tenens erstes enne nete EEEa EAE 19 Using the test client Isp client example esses enne nnns 20 Troubles
14. calhost in reply to MSG 2 reply with MSG 2 language ontology content ist Information EverythingPSC ist Information EverythingPSC ist Information EverythingPSC ist Information EverythingPSC isa Bluto Dog isa Dash VictoriasPet Dog isa HYP Dog 5968235 Dog isa GenericInstanceFn Dog Dog Statements of the form ist Information mt lt fact gt Indicate that fact fact holds in microtheory mt Typically you will be doing queries with regard to your own cases EverythingPSC is a problem solving context built into the OpenCyc ontology that treats the entire KB as one massive context This is good for spelunking through the KB but almost always is a bad idea otherwise The SME and MAC FAC functions are accessible programmatically via ask messages We discuss this in further detail below How to Use KQML to Store Storing is the act of inserting new facts into Case Mapper s KB You can create a store message by replacing the content field of the above packet with store fact fact context context The context is a required argument for specifying the logical context in the KB inside which the fact will be stored Contextualizing knowledge in the KB is important for maintaining the integrity of cases and for enabling reasoning and learning about rich human knowledge e g FrodoBaggins is a Hobbit in the MiddleEarthMt How to Use KQML to Retrieve Retrie
15. cept that is a superclass of your concept let s say super and include in your file genls new lt super gt You can have multiple genls statements concerning a concept in your file of course It is very useful to write comments about your new predicate You do this by the following statement comment new string describing new in your file These assertions will be used by the KB Browser so that you and potentially others can be reminded about your intent While SME will work without comments we have learned over the years that this is a very useful practice In Cyc derived KBs the convention is that relations start with lower case letters and are camel cased e g connectedTo functions start with an upper case letter and have Fn on the end e g FruitFn and attributes start with an upper case letter e g BiologicalSpecies Inthe SME Classics KB predicates functions and attributes are all lower cased with hyphens used to split terms instead of camel casing Case Mapper does not enforce these conventions but you will find it useful to stick with them when possible 4 You can learn more about emacs online at http www gnu org software emacs 16 8 2 2013 Generally you will find it useful to create a new meld file with the specifications for your new predicates and attributes separate from the meld files for whatever cases use them To indicate that these are background facts incl
16. e entity or collection name in the KB by typing part of its name and click Search KB Click the desired entity or collection Choose your desired case construction style discussed later and then click Load The case name is now added to the Base Case Target Case and Probe drop down lists in the uU Pp uN main window There are three case constructors that are built into Case Mapper currently They are MinimalCaseFn Retrieves the set of facts in the KB that mention the entity or collection CaseFn Like MinimalCaseFn except that it includes attribute information for every entity mentioned in the case 3 CleanCaseFn Like CaseFn except that redundant attributes are removed FIRE supports a number of other case constructors e g 6 but these three are useful for many purposes After you ve fetched cases you can run SME Alternatively you can use the case as your probe for MAC FAC Both of these operations are discussed with examples below 8 2 2013 Using SME from within Case Mapper We illustrate the use of SME via an example that uses the SME Classics smoke kb knowledge base We will fetch two cases from the knowledge base and then browse an analogical mapping between the cases These cases are from the Karla the Hawk stimulus set described in 3 How to Run SME 1 Open the smoke kb knowledge base Structure Mapping Engine SME 2 Fetch the case base 10 Base Case KBCeseFn base 10
17. eceived back from Case Mapper here MatcherFn 7 0 you can perform the following ask queries to obtain data about this SME analogy operation Just put the following lisp forms into the query argument of send test msg Match level queries numberOfMappings lt matcher handle gt x returns the total number of mappings computed here x 2 bestMapping matcher handle x returns a handle to the best mapping here x MappingFn 9 MatcherFn 7 0 Once you have a handle to a mapping you can use the following mapping level queries Mapping level queries numberOfCorrespondences mapping handle x returns the total number of correspondences in the mapping here x 11 5 The first integer refers to the order in which an SME was created during a session and the second integer refers to the number of times its base or target has been extended since SME supports incremental mapping In this example this is the 7 SME created during the session and no additional facts have been added to the base and target since their creation 21 8 2 2013 numberOfCandidateInferences lt mapping handle gt x returns the total number of candidate inferences in the mapping here x 3 structuralEvaluationScoreOf lt mapping handle gt x returns the structural evaluation score of the mapping here x 0 1657 correspondsInMapping lt mapping handle gt x y returns a list of x base and y ta
18. f the web browser The browser interface allows you to search for terms in the KB and manually remove or forget facts from cases in the KB Searching is easy enter a search term or the beginning of a search term i e ba in the SME Classics KB will bring up concepts such as bad bad tempered baker etc Warning Searching with a small search string e g ba in a large KB e g OpenCyc on a slow computer e g a laptop may take a very long time to complete It s better to search with a more specific term and use the links in the concept descriptions discussed below to navigate when possible Let s pick ball from the list of concepts using the SME Classics KB What you ll see looks something like this List All Explicit Cases search List All Case Libraries Edit KB Facts ball ball Smoke Test KB TACH5 all references El D zi B type Entity C QRG KBs CaseMapper KBs smoke kb Dynamic Cases MinimalCaseFn CaseFn Knowledge Base C QRG KBs CaseMapper KBs smoke kb Smoke Test KB TACH5 7 28 2013 Since ball is an entity the set of operations we can do on it is limited If you click on all references you should see something like this All References ball in sheat flow1 flow marble ball heat bar greater amount of heat marble amount of heat ball greater heat capacity marble heat capacity ball greater temperature marble temperature ball This tells y
19. hooting Case Mappel cccccccccccsssssesscecececessessaeeeceescesseeeaeeeeeessessesaeaeseeeessessesaeaeeeesesssessaeaneeesens 23 Testing Case Mapper unidad 23 RETEPONCES 24 8 2 2013 Appendix SME Parameters soii A tt da ada 25 Boolean Parameters iia ainia 25 NUOME FIC Paramete seenen m 26 Experimental Parameters ccccccssssssccececesseseneseeeeecessesesaeseeeeecesseseaaeseeececeseesaaaeeeeeessessesaaaeseeeessussesaaeees 26 8 2 2013 Overview Case Mapper provides an interface for using computational models of structure mapping theory 1 The Structure Mapping Engine SME 2 3 models analogical matching MAC FAC 4 models similarity based retrieval These are embedded in the FIRE reasoning engine 5 which provides a knowledge base infrastructure and various reasoning facilities Case Mapper is designed for cognitive scientists This means we assume some familiarity with knowledge representation predicate calculus and analogical reasoning but we do not assume any programming skills on your part Being a research software Case Mapper is a bit fussier than a commercial application but it is a software that our group uses daily in a variety of systems hence the internal operations are quite robust This manual describes how to operate Case Mapper Specifically e Use Knowledge Bases o Browse knowledge o Modify Delete existing knowledge o Add new knowledge e Use
20. ill list the candidate inferences for this mapping The base 10 ta 10 candidate inferences are facts about M ing 68 SME 2 temporal order of events and causality You can browse each ee candidate inference by clicking the button this will list Score 10 3780 the source and structural support for the inference You will Base KBCaseFn base 10 Target KBCaseFn ta 10 notice in this example expressions like skolem asset2 These are analogy skolems entities whose existence is Support Baseitem targetitem MHScore hypothesized by the candidate inference Notice that SME is Ara me ivan mefenristine noo 7 k 12 boris 1 0000 very conservative by default concerning candidate Me mj w A we jap efre imm inferences only inferring structure when supported by Xu jesse relational matches For some purposes e g building up a o fete model of a new domain where little is known about the wo ment target a more liberal criterion can be used by modifying the us we wwe pow Allow entity supported inferences SME parameter SME 3 candidate inferences Noreverse candidate inferences parameters are discussed in appendix 47 expression correspondences 30 functor correspondences Expression correspondences are facts that correspond Legend between the base and target cases You can browse these by ak Match Expression candidate Inference clicking
21. ion 4 In your IDE evaluate send test msg ask query isa x harmful This splices the lisp form ask query isa x harmful into the content argument of a new KQML message and sends it to Case Mapper The outgoing message and response message should be printed in the IDE window and the function should return the content from the response message which contains Case Mapper s response to the ask query Note that the term x is considered a variable by Case Mapper s internal reasoner since it is proceeded by a question mark As a context is not specified EverythingPSC all of KB is assumed to be the context 5 n your IDE evaluate the following send test msg store fact isa America country context cm test case This stores the fact isa America country in the KB in a microtheory contextual case called cm test case If the context cm test case did not exist beforehand Case Mapper creates it 6 Verify that the fact was properly stored by browsing the KB via the KB menu and searching for cm test case You ll notice that there is one fact in the microtheory the one we just stored programmatically using our KQML client 7 Retrieve the case by evaluating send test msg retrieve pattern isa America country context cm test case note that the context keyword is optional here and forget the fact by evaluating send test msg forget fact isa America country context cm test case Note that you
22. knowledge base in the world We do not use Cycorp s reasoning engine since our FIRE reasoning engine is optimized for analogical operations The subset of OpenCyc concepts we extracted currently contains just over 1 3M facts 58K concepts 14K relations and 3K functions We distribute this KB to support cognitive scientists who want to do larger scale experiments You will find that some operations are much faster with the SME classics KB although there is much more knowledge in the OpenCyc KB E QRO Case Mapper Beta oo o ES How to Open a Knowledge Base ree No KB is currently open Once Case Mapper is open your window should Structure Mapping Engine SME Base Case flo ai look similar to Figure 1 ae gt MAC FAC Reminding Open the KB menu and click Open KB Probe zr Case Library d Navigate to the directory where your KBs are stored by default it s C QRG KBs CaseMapper KBs under Windows rims Status No KQML server is active Port 4444 Start Server and for Linux it s lt CaseMapper gt Data KBs and select a Figure 1 Case Mapper is open for business 1 It is sometimes called the smoke KB because it is automatically constructed every time the FIRE regression test is run Regression tests are also known as smoke tests by analogy with seeing if a new piece of equipment smokes when it is first plugged in www opencyc org 8 2 2013 KB by se
23. l of similarity based retrieval MAC FAC Reminding im Probe KBCaseFn base 13 y Exclude probe from results The inputs to MAC FAC are 1 a probe which can PRO Rm rre be any existing or dynamically created case and Browse Library 2 a case library which is a set of persistent cases inthe KB You can use Case Mapper to select a probe and a case library and run MAC FAC transparently You can select a probe by fetching a case from the KB as discussed above In the following we discuss the processes of loading a case library and running MAC FAC Figure 9 MAC FAC in Case Mapper How to Select a Case Library A case library is a set of cases A single KB can contain many case Load Data from KB E libraries The KB must be open before selecting a case libraries To A 3 sys Search KB select a case library from the KB Corresponding data in KB SyslitCaseLibrary m 1 Under the File menu click Fetch Case then Fetch Case SyslitCaseLibrary12 SyslitCaseLibrary5 Library A pop up dialog will appear Figure 10 2 Locate the case library by typing part of the case library s name and click Search KB Note If you are using the smoke kb there are several pre existing case libraries as in Figure 1 0 Case Construction Style 3 Click the desired case library and click Load The selected case library will now be present in the Case Library drop down list within the MAC FAC interface For exam
24. lecting the directory of the KB you wish to use e g smoke kb or opencyc kb as in Figure 2 and then click OK 3 Case Mapper will display the current KB the Knowledge Base KB field Once a KB is open you can browse and edit the KB click Browse KB in the KB menu add new knowledge to the KB with meld files discussed below or fetch cases for use with SME and MAC FAC To change KBs click Close KB in the KB menu and open another as described above e Select a KB Directory How to Back up a knowledge base Selectthe directory of the knowledge base you wish to use gt My Documents Things can go wrong with any program and more so with lac mestirs amp 3 QRG KBs research software We strongly recommend backing up your KBs fem 02 opencyc kb before doing anything else with Case Mapper In the worst case a smoke kb amp D fixed NTFS Data b Network Machines you can always reinstall Case Mapper but that takes even longer Once you open a KB discussed next you can back it up for later use The command to use is on the KB menu Create KB Backup It will ask you for a directory to place the backup in Optionally enter a full path or one relative to CAORG KBslCaseMapperKBsl You can choose any directory you like e g a USB flash drive and Press OK to return the existing folder the backup dialog lets you type in the name of a new CAQRG KBsICasellapperKBs subdirectory along that path
25. location for the test result file The result file will contain the current SME parameter settings and other test statistics 23 8 2 2013 We request you to include the result file along with any bug report References 1 Gentner D 1983 Structure mapping A theoretical framework for analogy Cognitive Science 7 155 170 Falkenhainer B Forbus K and Gentner D 1989 The Structure Mapping Engine Algorithm and examples Artificial Intelligence 41 1 63 Forbus K Ferguson R and Gentner D 1994 Incremental structure mapping Proceedings of the Cognitive Science Society August Forbus K Gentner D and Law K 1995 MAC FAC A model of similarity based retrieval Cognitive Science 19 141 205 Forbus K Hinrichs T de Kleer Usher J J Klenk M Lovett A and Paritosh P unpublished The FIRE Manual Mostek T Forbus K and Meverden C 2000 Dynamic case creation and expansion for analogical reasoning Proceedings of AAAI 2000 Austin TX Falkenhainer B 1987 An examination of the third stage in the analogy process Verification based analogical learning In Proceedings of IJCAI 87 Los Altos Morgan Kaufmann 24 8 2 2013 Appendix SME Parameters CaseMapper provides an interface Edit SME Parameters for customizing SME by modifying the default parameters We recommend that you become very familiar with the SME algorithm before changing any of the default settings Edi
26. ments This is a psychologically important distinction In Cyc style knowledge bases attributes are implemented via collections Instances are related to collections via isa statements i e isa Nero Cat 8 2 2013 This says that Nero is a Cat i e the individual Nero is an instance of the collection Cat It is logically equivalent to the attribute form that SME expects i e Cat Nero Case Mapper automatically translates between these two forms as needed When you see a case in the KB browser you will see isa statements being used whereas when you see the same case in the SME or MAC FAC results browser you will see it in the attribute form For more information on SME including a discussion of its algorithm and more in depth explanation of the terms discussed above see 2 3 The Case Mapper distribution comes with two knowledge bases e SME Classics smoke kb directory This KB contains predicate specifications and cases from many of the early published SME and MAC FAC experiments It contains a minimal ontology as well as all of the stimuli from the Karla the Hawk experiments and other examples from the literature Experimenting with this KB is a great place to start e OpenCyc derived opencyc kb directory This KB contains contents we extracted from the OpenCyc knowledge base OpenCyc is a freely available version of the Cyc knowledge base which is currently the largest and most comprehensive formally represented
27. n a particular session with Case Mapper If you constrained SME as we do in the Modos ak Match A Expression Candidate Inference next example your constraints appear on this page as Hypothesis Detalls m petalis Explanation well Figure 6 Browsing SME Clicking the Base or the Target case names will let you browse the facts in each representation Clicking on any of the mappings will provide more information about that analogical mapping Note that SME may compute multiple mappings and in some cases no mappings These are not errors the number and content of mappings is a function of the facts in the 10 8 2 2013 base and target cases constraints placed on the mapping process discussed below and the parameters of SME discussed in the appendix Click on the mapping to browse it in more detail as shown in Figure 7 At the top is a score in this case 10 378 that reports the overall structural evaluation score of the selected mapping Note that this score is not normalized You will also notice a list of entity correspondences from the base and target Each correspondence has support which are expressions that provide evidence for the correspondence as well as a match hypothesis score computed from the support which reflects SME s confidence of the correspondence Candidate inferences are facts that may be true about the target based on the mapping between base and target Clicking the candidate inferences link w
28. o see the other capabilities of the KB Brower please switch to the OpenCyc KB by closing the SME Classics KB and then opening the OpenCyc KB Search for bas If you click on the BaseballDiamond concept you should see something similar to Figure 3 BaseballDiamond is a Collection i e an attribute in structure mapping terms as noted in the header and it s a subtype of i e it gen s to AthleticField You can click both Collection and AthleticField to learn more about both concepts this is an efficient manner of navigating the knowledge base To the left of the main frame you may browse all the other BaseballDiamond OpenCyc KB genls and specs of the concept as well as retrieve all ipe Cuco MS references of the concept in the entire KB Retrieving all isa in UniversalVocabularyMt ExistingObjectType references may take a very long time if there are a lot of SpatiallyDisjointObjectType them For example there are over a half million isa gene in UniversalVocabularyMt AthleticField statements in the KB so retrieving all of them is rather time t Knowledge Base C ORG planb kbs opencyc kb OpenCyc KB 9 18 2009 consuming and can even lead to time out errors see Troubleshooting Figure 3 Browsing a KB concept To get a deeper understanding of CyCL language used in Cyc we recommend you to visit http www cyc com and browse through the documentation How to Fe
29. om the target case to the base case Correspondence level queries correspondenceBaseltem lt correspondence handle gt x returns the base item of the given lt correspondence handle gt correspondenceTargetItem correspondence handle x returns the target item of the given lt correspondence handle gt Candidate inference level queries candidateInferenceContent ci handle x returns the fact form of the given ci handle in the mapping 22 8 2 2013 supportScoreOf lt ci handle gt x returns the support score of the corresponding candidate inference extrapolationScoreOf ci handle x returns the extrapolation score of the corresponding candidate inference candidateInferenceCorrespondences lt ci handle gt x returns a set of correspondence handles representing the correspondences match hypotheses that support the given candidate inference We have demonstrated how to use SME via Case Mapper s KOML server with a Common Lisp client however all of these queries can be performed by other programs written in other languages provided they use the same KQML message structure and query syntax as discussed above Troubleshooting Case Mapper When you start the KB Browser or try to browse SME or MAC FAC results for the first time you may get a Windows Firewall pop up dialog that asks you whether or not you want to allow Case Mapper to communicate It is important that
30. ored within that case You can define multiple cases in one file as long as each set of facts is set off by a separate case statement as Figure 13 illustrates Case Bobl isa Bob Dog relationExistsExists loves Person Dog animalTypeMakesSoundType Dog BarkingSound relationInstanceExists anatomicalParts GenericInstanceFn Dog Mouth Case Bob2 isa Bobby Cat relationExistsExists loves Person Cat relationInstanceExists anatomicalParts GenericInstanceFn Cat Mouth Figure 13 A simple meld file containing two cases Bob1 and Bob2 Each case contains a list of facts You can write a case file using any text editor that you are comfortable with just make sure to save your file as filename meld Three important points 1 3 It is important that your text editor produce plain text files often called ASCII files or txt files in Save As menus since formatted files like doc html or pdf will not be correctly handled by Case Mapper s load routines Your case must use concepts and relations that are defined in the knowledge base for SME and MAC FAC to work properly You can find appropriate pre existing concepts by using the KB browser If you need to define your own please see the next section To make editing a little easier you might want to use a text editor that will match your parentheses for you this means making sure that there are equal numbers of right and left 15
31. ou that in the SME Classics KB the entity ball only appears in the case sheat flow1 Cases in FIRE are implemented by what are called microtheories in the OpenCyc ontology Thus if you click on the name of the case you ll get something like this 8 2 2013 heat flow1 erani sheat flow1 all ref rli s type Microtheory facts in mt 4 analogy ctrl no genlMt gy ctrl no g js analogy ctrl genIMt in BaseKB Microtheory horn clauses no genIMt gt nene genlMt BaseKB UniversalVocabularyMt The isa statement is Cyc s way of doing attributes The isa can be read as in the BaseKB microtheory the statement isa sheat flow1 Microtheory is true When facts are translated into SME s format such isa statements become attribute statements e g Microtheory sheat flow1 Notice on the left that one can see the total size of the case some of them are quite large Analogy control predicates will be explained later On the right of every page is a short list of menu items The most immediately useful one is List All Case Libraries which lists the case libraries that are currently in the knowledge base You will see amongst them SyslitCaseLibrary SyslitCaseLibrary12 and SyslitCaseLibrary5 the others are regression tests for analogical generalization not yet supported in Case Mapper These are from the Karla the Hawk studies and if you drill down into them you will find a number of cases to experiment with T
32. ple as shown in Figure 9 we have chosen KBCaseFn base 13 as the probe and SysLitCaseLibrary as the case library You can browse the given case library by clicking Browse Library Figure 9 This will open up the KB browser and display the contents of the selected case library Figure 10 Case Library fetch dialog How to Run MAC FAC Once you have loaded and selected a probe and a case library discussed above 1 Run MAC FAC by clicking Do Reminding Figure 9 If you select Exclude probe from results MAC FAC will exclude the probe from the case library when performing similarity based reminding This is a good practice after all the most similar case to a probe is the probe itself 13 8 2 2013 Clicking Do Reminding will open a new browser window with a display similar to Figure 11 The browser window Figure 11 has two frames the lower one shows the results of MAC FAC The results of the MAC stage are shown as MAC Results listing the cases with the highest content vector dot product to the probe These cases are inputs to the FAC stage which uses SME to find the most similar case s The output s of the FAC stage are listed under MAC FAC Results The upper frame Figure 11 shows the browser interface for MAC FAC which allows you to perform additional operations and also to run MAC FAC again if necessary Using this interface you can a Select a case from the Case Library list
33. raints for the mappings that happen during FAC stage aad Entities morris prison obscenities walls boxcar copst man2 harmonica In Probe KBCaseFn base 13 Remove Selected Figure 12 FAC Constraints candidates chosen in MAC stage You can impose constraints on the comparisons using Set FAC Constraints Figure 12 The interface is similar to Set Constraints discussed before except that we can add constraints related to only the entities in the probe 14 8 2 2013 Extending a knowledge base For your own experiments you may need to write your own cases including defining new predicates relations and attributes This section explains how to do this How to Write a case file The file format supported by Case Mapper is the meld file all files end in the extension meld You can see some examples of meld files by looking in the samples folder that was included in your installation of Case Mapper Meld files are easy to write with only a few simple rules Each case must start with a case statement of the form Case casename where casename is the name used to identify your case It is wise to use the KB browser to check and see if the name you are thinking of is already in use Loading a meld file that refers to an existing case will add to not replace the facts involving that case Facts following a case statement are st
34. resemble Figure 5 3 Click Set Constraints 4 Select Ivan from the entities in the base and select Christine from the entities in the target 5 Select Selected entities must NOT correspond from the Available constraints dropdown list and click Add The constraint popup should look like Figure 8 6 Click Done then click Run SME and browse the resulting SME mappings The SME browser will display the constraint we just added in addition to any other constraints you ve added In this case the mapping suffered a score decrease due to the constraint If you browse the mapping you ll notice that Ivan and Christine are not included in the entity correspondences which also prevents many of the expression correspondences from participating in the mapping SME Constraints KBCaseFn base 10 KBCaseFn ta 10 b The Structure Mapping Engine can constrain its mappings in a number of ways These controls allow you to manipulate a subset of these constraints for the mappings between base and target Available constraints Selected entities must NOT correspond y Add Current constraints Ivan and Christine must NOT correspond Remove Selected Entities In Base KBCaseFn base 10 In Target KBCaseFn ta 10 profit high fate marriage high pleasure business Done Cancel Figure 8 Constraining SME 12 8 2 2013 Using MAC FAC from within Case Mapper MAC FAC is a mode
35. rget correspondences in the given mapping for example correspondsInMapping MappingFn 9 MatcherFn 7 0 planet electron hasCorrespondence lt mapping handle gt x returns a list of correspondence handles for all correspondences in the mapping for example MhFn 1 MatcherFn 7 0 Thisisinthe form MhFn lt index gt lt matcher handle gt where Mh stands for Match Hypothesis Once you have correspondence handles you can ask correspondence level queries Similarly the next two queries produces correspondence handles correspondenceForBaseItem lt mapping handle gt planet y returns a list of the form correspondenceForBaseItem lt mapping handle gt planet correspondence handle correspondenceForTargetItem mapping handle electron y returns correspondenceForTargetItem lt mapping handle gt electron lt correspondence handle gt candidateInferenceOf x lt mapping handle gt returns a list of candidateInferenceOf lt ci handle gt lt mapping handle gt items where each lt ci handle gt can be used in the below Candidate inference level queries These are analogical inferences from the base case to the target case reverseCandidateInferenceOf x lt mapping handle gt returns a list of reverseCandidateInferenceOf lt ci handle gt lt mapping handle gt items where each lt ci handle gt can be used in the below candidate inference level queries These are analogical inferences fr
36. server Importantly the KQML server interface of Case Mapper can communicate with client programs written in any programming language provided they can open sockets and send properly formatted KQML messages KQML based interaction with Case Mapper involves four types of interactions 1 Knowledge Base store 2 Knowledge Base retrieve 3 Knowledge Base forget and 4 ask Next we discuss how to start and stop the KQML server and we review KQML packet structure and specifics including each of these four interactions 17 8 2 2013 How to Start amp Stop the KQML Server KQML Server On the bottom of the Case Mapper pp Status No KQML server is active Port 4444 interface is a KQML Server panel This panel shows the status of the server and contains Port field for entering the server port of your server default is 4444 and a button for starting and stopping the server The status label will change based on the status of the server Once you have started a server you can use a client program written in the programming language of your choice to communicate with Case Mapper and use the knowledge base the built in reasoner and SME Basic KQML amp Packet Structure Client programs can be written in any language permitted that they can open a TCP IP socket to the computer running Case Mapper Case Mapper allows a subset the standard KQML interactions so we can review them in their entirety The following string is a valid KQ
37. t SME Parameters amp The Structure Mapping Engine has a number of parameters that alter its behavior These parameters may be tweaked and saved though we recommend using the default values Allow entity supported inferences 0 8 Greedy cutoff value Functor trickle down 3 Maximum GMaps Y Allow out of mapping contributions Local score for identical functors Block most out of mapping contributions Local score for identical functions Require overlap for merging y Trickle down multiplier Y Enforce 1 1 minimal ascension r Maximum local score for a MH Use global candidate inference support Minimal ascension multiplier Compute reverse inferences Use less greedy greedy merge Reload Defaults Done Cancel The descriptions of these parameters are given below Boolean Parameters e Allow entity supported inferences We believe that there are two criteria people use for generating candidate inferences The default criterion is conservative only producing candidate inferences when supported by relational correspondences However there is another criterion that we believe is used when little is known about the target i e allowing candidate inferences based on correspondences just involving entities When this Boolean flag is false SME uses the default conservative criterion When it is true it also allows candidate inferences to be generated using the more risky criterion In Case Mapper this
38. tch a Case A case is a set of facts Consequently a case can represent an experience a stimulus a collection of domain knowledge a theory a story or any other set of knowledge Pre existing cases may be fetched directly from your KB or new cases can be automatically constructed from portions of the knowledge 8 2 2013 base e g what the system knows about an entity say a person named Tom Hinrichs You can do so using an appropriate case constructor The KB must be open before fetching cases To fetch an existing case from the KB 1 Under the File menu click Fetch Case then Fetch Load Data from KB zl Existing Case A pop up dialog will appear Figure 4 Case Name 2 Locate the case by typing part of the case s name and pes sexe click Search KB 1 3 Click the desired case and click Load The case name is now added to the Base Case Target Case and Probe drop down lists in the main window The example in Figure 4 is using the SME Classics KB and the esu base stories are from the Karla the Hawk story set 4 Case Construction Style KBCaseFn hd One can also dynamically generate cases from the contents of Load Cancel the knowledge base To fetch an entity or collection case from the KB Figure 4 The case loader popup dialog 1 Under the File menu click Fetch Case then Fetch Entity Case or Fetch Collection Case A pop up dialog will appear Figure 4 Locate th
39. ude at the start of the file the declaration in microtheory UniversalVocabularyMt This will make your new information accessible from every case How to Load a case file into your knowledge base If you have knowledge in a meld file that you want to use with SME or MAC FAC you must load it into your KB To load a meld file into your KB first ensure that you have a KB open and then choose Load File gt Import meld file from the File menu How to Test a case file Meld File Report x B gt O D 127 00 1 rbrowse meld reporthtmi id WARO Google Bookmark ORG Wiki CJ Announcements In Critical thinking web Automated Reasoni 1 Watch Person of Int EE Sign Off Magnum Air Servi Bid The Scientific Method 7j Other bookmarks search Meld File Report Smoke Test KB SUBU PC C QRG KBs CaseMapper KBs smoke kb Assertion in microtheory Bobi Problem Could not find a predicate definition for in microtheory Assertion relationExistsExists loves Person Dog Problem Could not find an arity for predicate relationExistsExists Try adding the fact arity relationExistsExists 3 Assertion animal TypeMakesSoundType Dog BarkingSound Problem Could not find an arity for predicate animalTypeMakesSoundType Try adding the fact arity animalTypeMakesSoundType 2 Assertion relationInstanceExists anatomicalParts GenericInstanceFn Dog Mouth Problem Could not find an arity for predicate relationinstanceExists Try
40. ving is the act of finding facts within Case Mapper s KB You can create a retrieve message by replacing the content field of the above packet with retrieve pattern query context context The context is an optional argument for specifying the logical KB context to perform the retrieve How to Use KQML to Forget Forgetting is the act of removing a fact from a logical context in Case Mapper s KB Because knowledge is contextualized forgetting a fact in one context will not affect the presence of a fact in another context You can create a forget message by replacing the content field of the above packet with forget fact fact context lt context gt The context is a required argument for specifying 19 8 2 2013 the logical context in the KB inside which the fact will be forgotten If the fact is not present in the given context no action will be taken Using the test client lsp client example This section assumes you have a Common Lisp IDE installed Also note that this section uses certain cases within the smoke kb but the same methods apply for any case in any Case Mapper compatible knowledge base Start Case Mapper open the smoke kb and start your KQML server on some port Compile and load the test client Isp file 3 Inyour IDE evaluate make test client localhost 4444 but substitute 4444 for the port you chose This creates and binds a persistent client object for this server port combinat
41. which should be used Importantly La cone the backup process deletes anything that is already in the Figure 2 Opening a KB directory that you give it to avoid KB corruption So unless you are replacing a previous backup creating a new directory is almost always what you want to do Should you need to use your backup the command Restore KB from Backup also on the KB menu can be used to replace the contents of a KB with the backup The SME Classics KB is tiny requiring only around 10 MB of storage The OpenCyc derived KB requires less than 2GB of storage How to Close Case Mapper When you are through with Case Mapper we recommend that you close the KB you were working with i e KB Close KB and then exit the program in the usual way File Quit We have set things up so that the KB should close automatically if you just quit though How to Browse a Knowledge Base Case Mapper enables you to browse the knowledge base in a web browser After you open a KB the menu command Browse KB in the KB menu will automatically open your browser to the correct address Sometimes this process can be cranky since communicating with browsers from other 3 We routinely use Internet Explorer Chrome and Firefox We have not systematically tested our code with other web browsers 8 2 2013 software isn t always smooth If the browser contents haven t changed within 30 seconds press the button again after starting a fresh copy o
Download Pdf Manuals
Related Search