n1 = '1000'
n2 = '100000'
strong_ext = '_chi2_strong_' + vectorizer
weak_ext = '_chi2_weak_' + vectorizer
train_features, train_labels = load_data.load_feature_data(0, test_train='train')
len_train= len(train_labels)
test_features, test_labels = load_data.load_feature_data(0, test_train='test')
all_labels = np.append(train_labels, test_labels)
## Do better Normalization on Customer Features
all_features = np.vstack((train_features, test_features))
float_feats = [[float(i) for i in row] for row in all_features] # Turn values to floating point
new_feats = preprocessing.normalize(float_feats, norm='l1', axis=0)
train_features = new_feats[:len_train]
test_features = new_feats[len_train:]
### Done Normalizing
train_email_features, test_email_features = load_data.load_email_data(0, vectorizer, stemmer=stem, vectorizer=v2)
train_subject_features, test_subject_features = load_data.load_subject_data(0, vectorizer, stemmer=stem, vectorizer=v2)
train_email_features_n1, test_email_features_n1 = load_data.load_email_data(0, n1+v2, stemmer=stem, vectorizer=v2)
train_subject_features_n1, test_subject_features_n1 = load_data.load_subject_data(0, n1+v2, stemmer=stem, vectorizer=v2)
train_email_features_n2, test_email_features_n2 = load_data.load_email_data(0, n2+v2, stemmer=stem, vectorizer=v2)
train_subject_features_n2, test_subject_features_n2 = load_data.load_subject_data(0, n2+v2, stemmer=stem, vectorizer=v2)
train_email_strong_features, test_email_strong_features = load_data.load_email_data(0, strong_ext, stemmer=stem, vectorizer=vectorizer)
train_subject_strong_features, test_subject_strong_features = load_data.load_subject_data(0, strong_ext, stemmer=stem, vectorizer=vectorizer)
train_email_weak_features, test_email_weak_features = load_data.load_email_data(0, weak_ext, stemmer=stem, vectorizer=vectorizer)
train_subject_weak_features, test_subject_weak_features = load_data.load_subject_data(0, weak_ext, stemmer=stem, vectorizer=vectorizer)
logging.info("All Data Loaded")
for t in ['email']:
for s in ['n2']:
if t == 'email':
if s =='strong':
vectorizers = ['TfidfVectorizer', 'HashingVectorizer']
stemmers = ['RegexpStemmer', 'LancasterStemmer', 'PorterStemmer']
for vectorizer in vectorizers:
for stem in stemmers:
train_features, train_labels = load_data.load_feature_data(0, test_train='train')
len_train= len(train_labels)
test_features, test_labels = load_data.load_feature_data(0, test_train='test')
all_labels = np.append(train_labels, test_labels)
## Do better Normalization on Customer Features
all_features = np.vstack((train_features, test_features))
float_feats = [[float(i) for i in row] for row in all_features] # Turn values to floating point
new_feats = preprocessing.normalize(float_feats, norm='l1', axis=0)
train_features = new_feats[:len_train]
test_features = new_feats[len_train:]
### Done Normalizing
train_email_features, test_email_features = load_data.load_email_data(0, vectorizer, stemmer=stem, vectorizer=vectorizer)
train_subject_features, test_subject_features = load_data.load_subject_data(0, vectorizer, stemmer=stem, vectorizer=vectorizer)
logging.info("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")
logging.info("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")
logging.info("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")
logging.info("VECTORIZER = %s" % vectorizer)
logging.info("STEMMER = %s" % stem)
logging.info("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")
logging.info("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")
logging.info("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")
logging.info("All Data Loaded")
for t in ['both', 'all', 'subject', 'email', 'normal']:
if t == 'email':
trainer = train_email_features
observed = classifier.classify(feats)
testsets[observed].add(i)
print 'accuracy:', nltk.classify.util.accuracy(classifier, testfeats)
print 'pos precision:', nltk.metrics.precision(refsets['pos'], testsets['pos'])
print 'pos recall:', nltk.metrics.recall(refsets['pos'], testsets['pos'])
print 'neg precision:', nltk.metrics.precision(refsets['neg'], testsets['neg'])
print 'neg recall:', nltk.metrics.recall(refsets['neg'], testsets['neg'])
classifier.show_most_informative_features()
def best_bigram_word_feats(words, score_fn=BigramAssocMeasures.chi_sq, n=400):
bigram_finder = BigramCollocationFinder.from_words(words)
bigrams = bigram_finder.nbest(score_fn, n)
return dict([(ngram, True) for ngram in itertools.chain(words, bigrams)])
if __name__ == "__main__":
train_subject, test_subject = load_data.load_email_data(0, extension='unVectorized', stemmer='PorterStemmer', vectorizer='unVectorized')
train_email, test_email = load_data.load_email_data(0, extension='unVectorized', stemmer='PorterStemmer', vectorizer='unVectorized')
train_features, train_labels = load_data.load_feature_data(0, test_train='train')
test_features, test_labels = load_data.load_feature_data(0, test_train='test')
train_email_features = [email.split(' ') for email in train_email]
test_email_features = [email.split(' ') for email in test_email]
train_subject_features = [subject.split(' ') for subject in train_subject]
test_subject_features = [subject.split(' ') for subject in test_subject]
for email_group in ['email', 'subject']:
for wordType in ['single', 'bigram', 'trigram']:
logging.info("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")
logging.info("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")
logging.info("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")
logging.info("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")
logging.info("Getting Word Stuff for %s and %s" % (email_group, wordType))
if email_group == 'email':
# trainer = train_features
# test = test_features
# split_train, split_labels = segment_for_even_distribution(trainer, train_labels)
# logging.info("============================================================")
# logging.info("Grid Search SVM on Customer Info")
# svmTrainandPrintWithGridSearch(split_train, split_labels, test_email_features, test_labels)
# #SPlit up the data to better parts
# for stem in ['PorterStemmer', 'RegexpStemmer', 'LancasterStemmer']:
# for vectorizer in ['TfidfVectorizer', 'HashingVectorizer']:
for n in [100, 1000, 10000, 100000, 1000000, 10000000, 100000000]:
stem = "RegexpStemmer"
vectorizer = "HashingVectorizer"
logging.info("Loading Data for %s, %s, %i:" % (stem, vectorizer, n))
train_email_features, test_email_features = load_data.load_email_data(
0, str(n) + vectorizer, stemmer=stem, vectorizer=vectorizer, num_features=n
)
train_subject_features, test_subject_features = load_data.load_subject_data(
0, str(n) + vectorizer, stemmer=stem, vectorizer=vectorizer, num_features=n
)
for textType in ["both"]:
logging.info("Building Model for %s " % textType)
if textType == "email":
train, test = train_email_features, test_email_features
elif textType == "subject":
train, test = train_subject_features, test_subject_features
elif textType == "both":
train = scipy.sparse.hstack([train_subject_features, train_email_features])
test = scipy.sparse.hstack([test_subject_features, test_email_features])
train = scipy.sparse.csr_matrix(train)
test = scipy.sparse.csr_matrix(test)
