Peek at the modelSentence to see its labels -- for training only! @param sentence @param modelSentence

Adds into labelScores

Training-only add-in loglik gradient (direction of higher likelihood)

Copies coefs from sourceModel into destModel. For observation features, only copies features that exist in both. (Therefore if a feature exists in des

This needs forward-backward I think @return dim: (T x K) posterior marginals at each position

Twitter text comes HTML-escaped, so unescape it. We also first unescape &'s, in case the text has been buggily double-escaped.

assume mSent's labels hold the tagging. @param lSent @param mSent @param inputLine -- assume does NOT have trailing newline. (default from java's re

lambda_2 (1/2) sum (beta_j)^2 + lambda_1 sum |beta_j| our OWLQN seems to only want the first term

Specify a callback for printing the weights after each iteration of optimization.

"foo bar " => "foo bar"

Assume 'text' has no HTML escaping.

Run the tokenizer and tagger on one tweet's text.

vit[t][k] is the max probability such that the sequence from 0 to t has token t labeled with tag k. (0<=t<T) bptr[t][k] gives the max prob. tag of

@return dim T array s.t. labelScores[t]+=score of label prior followed by label t