README.txt (this will be a separate file later)
This project requires Python 2.6 and the nltk package available at http://www.nltk.org/

ngram.py
Unigram and Bigram objects are constructed by either setting the filename flag or the text_string
flag to the desired data. The unk flags turns on unknown word handling.
For Unigrams, smoothed can be set to either True or False.
For Bigrams, smoothed can be set to NONE, LAPLACE, or GOOD_TURING.

The functions on Unigram.get_frequencies() and Bigram.get_frequencies() returns all the ngram frequencies.
get_probabilities() returns all the probabilities.
Unigram.get_probability(word) and Bigram.get_probability(first, second)
return individual ngram probabilities and accounts for unknown words.
generate_sentence() will print out a randomly generated sentence.


authorPredictor.py
This file runs the author predictor algorithm.
At the very bottom of the file in main, set up the call to email_prediction with the desired option and file paths.
The options are as follows
- singletons: uses only singleton unigrams to compute perplexity.
- remove_punctuation: removes punctuation marks, except for . and ! and ? from emails.
- farmer_correction: Makes algorithm more sensitive towards prediction Òfarmer-dÓ
- kaggle: includes the test data for submission to kaggle. Output is written to submission.csv.
The perplexity functions are also in this file.
perplexity2() takes in a test and model unigram or bigram objects.
test_perplexity() runs perplexity2() on test and train files
(their paths are at the top of the function) using different smoothing techniques.