TODO

o Improve ClassLm
o Add LMServer
o Add __str__ and/or __repr__ for classes
o Make sure numpy.array can be used in place of array.array
x Add CacheLM
x Add more discounting methods
x Add maxent.lm module
x Improve Makefile to run tests with 'make test'
x Add docstring docs
x Debug ngram.stats iterator [Use Vocab_None as end-of-seq symbol!!!]
x Add more tests; need class TestNgramLM
x Add countFile() to ngram.stats; note limitVocab
x Add make_test() helper function
x Add a class to hold discount params
x Improve estimate() with more smoothing methods
x Add a Sphinx set up
x Add test() to ngram.lm
x Refactorize Discount
x Test Lm.read() and Lm.write() arpabo format
x Add read() and write() for Discount
x Compare results with directly invoke srilm
x Add iterator for Lm
x Improve discount module with duck-typing