A ternary search tree is a hybrid between a binary tree and a digital search tree (trie). Keys are limited to strings. A data value of type char is stored in each leaf node. It can be used as an index (or pointer) to the data. Branches that only contain one key are compressed to one node by storing a pointer to the trailer substring of the key. This class is intended to serve as base class or helper class to implement Dictionary collections or the like. Ternary trees have some nice properties as the following: the tree can be traversed in sorted order, partial matches (wildcard) can be implemented, retrieval of all keys within a given distance from the target, etc. The storage requirements are higher than a binary tree but a lot less than a trie. Performance is comparable with a hash table, sometimes it outperforms a hash function (most of the time can determine a miss faster than a hash).
The main purpose of this java port is to serve as a base for implementing TeX's hyphenation algorithm (see The TeXBook, appendix H). Each language requires from 5000 to 15000 hyphenation patterns which will be keys in this tree. The strings patterns are usually small (from 2 to 5 characters), but each char in the tree is stored in a node. Thus memory usage is the main concern. We will sacrifice 'elegance' to keep memory requirements to the minimum. Using java's char type as pointer (yes, I know pointer it is a forbidden word in java) we can keep the size of the node to be just 8 bytes (3 pointers and the data char). This gives room for about 65000 nodes. In my tests the English patterns took 7694 nodes and the German patterns 10055 nodes, so I think we are safe.
All said, this is a map with strings as keys and char as value. Pretty limited!. It can be extended to a general map by using the string representation of an object and using the char value as an index to an array that contains the object values.
@author cav@uniscope.co.jp
|
|
|
|
|
|
|
|
|
|
|
|
|
|