def main(): # Load configuration from file config = Configuration(config_file='classify.conf') try: config.load_configuration() config_data = config.get_configuration() except: print "Error loading configuration file." print "Classifier aborting." raise #config.display_configuration() print config #sys.exit() myfolds = config.get_folds() correctness = 0 #Preporcessor: tokenizer, stemmer, etc. prep_lower = config_data['lower'] prep_stem = config_data['stem'] prep_pos = config_data['pos'] prep_ngram = config_data['ngram'] prep = Preprocessor(pattern='\W+', lower=prep_lower, stem=prep_stem, pos=prep_pos, ngram=prep_ngram) for myfold in myfolds: ev = Evaluation(config=config, fold=myfold) if config_data['classifier'] == 'rocchio': ml = Rocchio(verbose=VERBOSE, fold=myfold, config=config, ev=ev) elif config_data['classifier'] == 'knn': ml = KNN(verbose=VERBOSE, fold=myfold, config=config, ev=ev) else: ml = NaiveBayes(verbose=VERBOSE, fold=myfold, config=config, ev=ev) training(config, myfold, ml, prep ) ml.do_padding() ml.calculate_training_data() #r.display_idx() ml.diagnose() testing(config, myfold, ml, ev, prep) k = config_data['k'] results = ev.calculate(review_spam=True, k=k) print 'Accuracy for fold %d: %s' % (myfold, results) correctness += results print "\nAverage accuracy for all folds:", correctness / len(myfolds)
def main():
# Load configuration from file
config = Configuration(config_file='/home/huma/Downloads/irlib-0.1.1/irlib/classify.conf')
try:
config.load_configuration()
config_data = config.get_configuration()
except:
print("Error loading configuration file.")
print("Classifier aborting.")
raise
# config.display_configuration()
print(config)
# sys.exit()
myfolds = config.get_folds()
correctness = 0
# Preporcessor: tokenizer, stemmer, etc.
prep_lower = config_data['lower']
prep_stem = config_data['stem']
prep_pos = config_data['pos']
prep_ngram = config_data['ngram']
prep = Preprocessor(pattern='\W+', lower=prep_lower, stem=prep_stem, pos=prep_pos, ngram=prep_ngram)
for myfold in myfolds:
ev = Evaluation(config=config, fold=myfold)
if config_data['classifier'] == 'rocchio':
ml = Rocchio(verbose=VERBOSE, fold=myfold, config=config, ev=ev)
elif config_data['classifier'] == 'knn':
ml = KNN(verbose=VERBOSE, fold=myfold, config=config, ev=ev)
else:
ml = NaiveBayes(verbose=VERBOSE, fold=myfold, config=config, ev=ev)
training(config, myfold, ml, prep)
ml.do_padding()
ml.calculate_training_data()
# r.display_idx()
ml.diagnose()
testing(config, myfold, ml, ev, prep)
k = config_data['k']
results = ev.calculate(review_spam=True, k=k)
print('Accuracy for fold %d: %s' % (myfold, results))
correctness += results
print("Average accuracy for all folds:", correctness / len(myfolds))
def training(self, first_n_files=10000): self.ev = Evaluation() self.ml = Rocchio(verbose=VERBOSE, fold='n/a', config=self.config, ev=self.ev) self.parse_files(fold="sgms/TRAIN.jdb", mode='training', first_n_files=first_n_files, ml=self.ml, config=self.config, prep=self.prep) self.ml.do_padding() self.ml.calculate_training_data() self.ml.diagnose()
class Classify():
def __init__(self, printMessage=None):
self.prep = SimpleAnalyzer(expression=r"[A-Za-z]*", gaps=False)
self.printMessage = printMessage
def setConfig(self):
self.config = Configuration(config_file='classify.conf')
try:
self.config.load_configuration()
config_data = self.config.get_configuration()
except:
self.printMessage("Error loading configuration file.")
self.printMessage("Classifier aborting.")
raise
# config.display_configuration()
self.printMessage(self.config)
# Parse not any more than the first_n_files in folder
# @ml: Object for our classifier class (Rocchio, kNN, etc)
# @config: Our configuration class (class as in OOP not ML)
# @prep: Preprocessor class; tokenizers, stemmers, etc.
def parse_files(self, fold=1, mode = "training", first_n_files = 500, ml=object, config=object, prep=object):
config_data = config.get_configuration()
fd = open(fold, 'r')
file_data = fd.read()
objs = json.loads(file_data)
counter = 0;
length = len(objs)
if length > first_n_files :
length = first_n_files
self.printMessage("%s data length: %d" % (mode,length))
for i in objs:
if counter > first_n_files:
break
counter += 1
if counter % 100 == 0:
self.printMessage(counter)
doc_id = objs[i][unicode('id')]
# terms = prep.ngram_tokenizer(text=objs[i][unicode('content')])
terms = [token.text for token in prep(objs[i][unicode('content')]) if token.text != "" ]
# self.printMessage(terms)
for class_name in objs[i][unicode('topics')]:
if mode == 'training':
ml.add_doc(doc_id = doc_id, doc_class=class_name, doc_terms=terms)
else:
# Class known from filename
ml.add_query(query_id = doc_id, query_class=class_name, query_terms=terms)
fd.close()
# Let's do some workout now on all folders but one
def training(self, first_n_files=10000):
self.ev = Evaluation()
self.ml = Rocchio(verbose=VERBOSE, fold='n/a', config=self.config, ev=self.ev)
self.parse_files(fold="sgms/TRAIN.jdb", mode='training', first_n_files=first_n_files, ml=self.ml, config=self.config, prep=self.prep)
self.ml.do_padding()
self.ml.calculate_training_data()
self.ml.diagnose()
# Let's test on the remaining folder
def testing(self, first_n_files=10000):
self.parse_files(fold="sgms/TEST.jdb", mode='testing', first_n_files=first_n_files, ml=self.ml, config=self.config, prep=self.prep)
self.ml.compare_queries()
def result(self):
results = self.ev.calculate(review_spam=True, k=self.config.get_configuration()['k'])
self.printMessage(results)