@return The number of individual words in the vocabulary

@return

Calculates the distribution of an entire Cluster. Similar to calculating a centroid with k-means. @return a "Map" vector

Wipes all data apart from the parent vocabulary. Done this way instead of destructing the object altogether in order to keep a constant number of clus

@return The smoothed distribution of the word in (K+1)^2 space.

@param threshold - Typically the "Rare word threshold" @return The number of words below a certain threshold.

@param a A cluster @param b Another cluster @return the KLD: D(a||b)

INNEFICIENT - Generates each time @return

@param i The word index @return True if the word is registered as updated

Updates the word

To be called rarely. Builds the n-gram database and vocabulary

MAIN ALGORITHM METHOD

To be updated at every turn.

Step 1 of the Algorithm

Resets the state of updated words

Updates are wiped every iteration, so updating these is important. @param c

@param i

Check to see if any distance between two clusters falls below epsilon (defined above), then update parenthood, add everything to c1, and reset c2.