JavaNLP Meetings Notes 11/19/2003

Updates:
 
Jenny:
 
Wrote a Quasi-Newton optimization algorithm (QNMinimizer) that works just like CGMaximizer, but about 3x as fast!  Currently investigating why it’s not 10x as fast.
 
Kristina:
 
ObjectBank: Almost there.  Jenny will bring the revised API to this week’s meeting, and hopefully, we can begin implementation. 
 
/*
-         Vends iterators over the Collection.  
-         The iterator returns Readers, not the items in the source Collection.  
-         Since it vends iterators rather than being an iterator itself, it is resettable
*/
ReaderIteratorFactory: 
 
/* takes a Collection of URLs, Files, Strings (can be mixed) */
            ReaderIteratorFactory(Collection c)      
 
/* returns an iterator (internally defined) over the collection, the iterator will check the runtime type of the Collection item, and return
iterator() a StringReader if it’s a String, FileReader if it’s a file, etc. */
 
            /* adds items to the Collection */
            add(Object/Collection) 
 
/* 
-         Vends iterators, which can process a given Reader in some way to produce Objects
-         Subclass this to process different kinds of data
*/
ObjectIteratorFactory 
 
            Predicted subclasses:
/* produces Documents by tokenizing on specified delimiters */                         
                        DocumentIteratorFactory                                  
DocumentIteratorFactory(String delimiters)
                                    DocumentIteratorFactory(Tokenizer tokenizer)
                        /* produces Trees from the input Reader */
                        TreeIteratorFactory
 
/* an iterator where remove() throws an UnsupportedOperationException */
AbstractIterator
 
/* Takes a ReaderIteratorFactory and an ObjectIteratorFactory and combines them to produce an iterator that vends objects */
{Disk/Memory}ObjectBank