04/11/2006 AL Group Meeting * Agenda - Go over Andy's draft of API code. - Go over Kevin & Sophie's bootstrapper - Discuss Rosie Jones thesis. Divide AL derivation & coding tasks (diff. versions, hopefully). 1. Kevin's main-loop: Simple Boot(relation,seed1, seed2,….seedn) AL=NewAcitveLearner(relation) For(1…n) AL.addRule(seed[], seed, relation) While there are unsearched rules { Rules=A.getRankedRuleList(relation) For each rule in rules Extract entities , and for each entity Al.addEntityCont(rule, entity, counts) Entities=AL.getRankedEntityList(relation, 5) Extract rules and for each unsearched entity AL.addRule(rule, entities, relation) HandleQuestions() } Entities=AL.getHighProbEntities(relation) Print them 2. The Question object Questions are of the form: Q: Is Poland a -Nation -None of the above. -I don’t know Q: Are (US, Poland) -allies -enemics -None of the above. -I don't know We threw out: Q. Is Poland a nation? -Yes -No -I don't know So the question object should contain the entity(pair) and something like the list of relations. And a print method? 3. Rest of the API // List GetQuestions() Question GetQuestion() void AnswerQuestion( int ) Function() Fact(relation, entity) RankQuestion() ScoreEstimator() *Things we threw out RegisterQuestion(question, learnObj) UI: Void RegisterQuestion(q, ) Void Handle Outstanding Question