def compare_classifiers(test_set, full_training_labeld_features, step, max_size): """ compare Naive Bayes with MaxEnt on different training set sizes """ nb_acc = [] me_acc = [] for size in range(step, max_size+1, step): print "creating trainig set of size", size training_set = create_even_training_set(size, full_training_labeld_features) print "train NBSentimentClassifier" nb_classifier = NBSentimentClassifier().train(training_set) nb_acc.append(nb_classifier.test_accuracy(test_set)) print "train MaxEntSentimentClassifier" me_classifier = MaxEntSentimentClassifier().train(training_set) me_acc.append(me_classifier.test_accuracy(test_set)) return nb_acc, me_acc
""" iterator that splits a list into num_chunks chunks. truncates remainder """
slice_len = len(alist)/num_chunks
for x in xrange(0, num_chunks):
yield alist[x*slice_len:(x+1)*slice_len]
def combine_dicts(a, b, op=None):
op = op or (lambda x, y: x + y)
return dict(a.items() + b.items() +
[(k, op(a[k], b[k])) for k in set(b) & set(a)])
comm = MPI.COMM_WORLD
rank = comm.Get_rank()
size = comm.Get_size()
classifier = NBSentimentClassifier().load_model()
if(rank == 0):
if len(sys.argv) > 1:
csvFile = sys.argv[1]
keyword = sys.argv[2]
else:
csvFile = 'trainingandtestdata/testdata.csv'
tweetlist = loadTwitterCSV(csvFile)
#tweetlist = loadTwitterCSV('trainingandtestdata/training.1600000.processed.noemoticon.csv')
tweetlist = chunked(tweetlist, size)
else:
tweetlist = None # tweetlist must be defined
args = parser.parse_args()
print "creating feature sets..."
tweetlist = loadTwitterCSV('trainingandtestdata/testdata.csv')
labeld_features = label_feats_from_tweets(tweetlist)
#training_set, test_set = split_label_feats(labeld_features)
tweetlist = loadTwitterCSV('trainingandtestdata/training.1600000.processed.noemoticon.csv')
training_set = label_feats_from_tweets(tweetlist)
training_set, garbage = split_label_feats(training_set, 1.0)
test_set, garbage = split_label_feats(labeld_features, 1.0)
print "training set length: %i test set length: %i" % (len(training_set), len(test_set))
#print prettifyFeatureSet(test_set)
if args.algo == 'nb':
classifier = NBSentimentClassifier().train(training_set)
print "training NaiveBayes classifier..."
else:
classifier = MaxEntSentimentClassifier().train(training_set)
print "training MaxEnt classifier..."
print "calculating accuracy..."
print 'accuracy:', classifier.test_accuracy(test_set)
#classifier.show_most_informative_features(30)
classifier.save_model()
# load a serialized trained classifier
#classifier = NBSentimentClassifier().load_model()
#classifier = MaxEntSentimentClassifier().load_model()