Investigate Double-Array Dawg

[jeanbern]

Expected outcome is that it will take a bit less space but run slower.

Could just implement both and leave them available for injection based on user preference.

Trade-offs:

• DADAWG uses less memory. Maybe somewhere on the order of 30% less.
• DADAWG can traverse down a know branch in O(1) while DAWG needs O(log |alphabet|).
• DAWG can traverse all branches in O(|branches|) while DADAWG requires O(|alphabet|) in all cases.
• Creation of DADAWG might be slower. Depends on how good of a space-filling algorithm I can create.

[jeanbern]

Just as a refresher for what this means:
DADAWG has two arrays. One representing the letter you came from (for verification) and another representing states. To go from state x and traversing through a given letter. You would look at states[x + letter] for the new state address. Then you double check that verification[x+letter] == letter. This is faster the the case where you check if a word exists, but a bit slower when doing near-neighbor detection.
It requires a packing methods to efficiently fit states next to each other with no overlap of state + letter. But greedy is good enough apparently [http://citeseerx.ist.psu.edu/viewdoc/citations;jsessionid=F9F10F8B3A18B7B4D8370A12C23B80F9?doi=10.1.1.56.5272]