def iterator(dataset_fn):
"""
Provides an iterator of parsed documents from the 20 Newsgroups dataset.
:param dataset_fn: Path to Newsgroups dataset archive file.
:type dataset_fn: unicode|str
:rtype : generator
"""
ng = datasets.fetch_20newsgroups()
for article, group, target, filename in zip(
ng['data'], [ng['target_names'][x] for x in ng['target']],
ng['target'], ng['filenames']):
article = twenty_newsgroups.strip_newsgroup_header(article)
article = twenty_newsgroups.strip_newsgroup_footer(article)
article = twenty_newsgroups.strip_newsgroup_quoting(article)
doc_id = os.path.basename(filename)
yield {
'doc_id': doc_id,
'article': article,
'group': group,
'target': target,
'filename': filename
}
def transform(self, posts):
features = np.recarray(shape=(len(posts), ),
dtype=[('subject', object), ('body', object)])
for i, text in enumerate(posts):
headers, _, bod = text.partition('\n\n')
bod = strip_newsgroup_footer(bod)
bod = strip_newsgroup_quoting(bod)
features['body'][i] = bod
prefix = 'Subject:'
sub = ''
for line in headers.split('\n'):
if line.startswith(prefix):
sub = line[len(prefix):]
break
features['subject'][i] = sub
return features
def transform(self, posts):
features = np.recarray(shape=(len(posts),),
dtype=[('subject', object), ('body', object)])
for i, text in enumerate(posts):
headers, _, bod = text.partition('\n\n')
bod = strip_newsgroup_footer(bod)
bod = strip_newsgroup_quoting(bod)
features['body'][i] = bod
prefix = 'Subject:'
sub = ''
for line in headers.split('\n'):
if line.startswith(prefix):
sub = line[len(prefix):]
break
features['subject'][i] = sub
return features
def iterator(dataset_fn):
"""
Provides an iterator of parsed documents from the 20 Newsgroups dataset.
:param dataset_fn: Path to Newsgroups dataset archive file.
:type dataset_fn: unicode|str
:rtype : generator
"""
ng = datasets.fetch_20newsgroups()
for article, group, target, filename in zip(ng['data'], [ng['target_names'][x] for x in ng['target']],
ng['target'], ng['filenames']):
article = twenty_newsgroups.strip_newsgroup_header(article)
article = twenty_newsgroups.strip_newsgroup_footer(article)
article = twenty_newsgroups.strip_newsgroup_quoting(article)
doc_id = os.path.basename(filename)
yield {'doc_id': doc_id, 'article': article, 'group': group, 'target': target, 'filename': filename}
def transform(self, posts):
# construct object dtype array with two columns
# first column = 'subject' and second column = 'body'
features = np.empty(shape=(len(posts), 2), dtype=object)
for i, text in enumerate(posts):
headers, _, bod = text.partition("\n\n")
bod = strip_newsgroup_footer(bod)
bod = strip_newsgroup_quoting(bod)
features[i, 1] = bod
prefix = "Subject:"
sub = ""
for line in headers.split("\n"):
if line.startswith(prefix):
sub = line[len(prefix):]
break
features[i, 0] = sub
return features
def transform(self, posts):
# construct object dtype array with two columns
# first column = 'subject' and second column = 'body'
features = np.empty(shape=(len(posts), 2), dtype=object)
for i, text in enumerate(posts):
headers, _, bod = text.partition('\n\n')
bod = strip_newsgroup_footer(bod)
bod = strip_newsgroup_quoting(bod)
features[i, 1] = bod
prefix = 'Subject:'
sub = ''
for line in headers.split('\n'):
if line.startswith(prefix):
sub = line[len(prefix):]
break
features[i, 0] = sub
return features
def fetch_sraa(
DOWNLOAD_URL='http://people.cs.umass.edu/~mccallum/data/sraa.tar.gz',
dataset_home='../datasets/SRAA',
dopickle=False):
"""
Fetchs a version of Reuters21578 for cross-domain adaptation, as defined in:
Dai, W., Xue, G. R., Yang, Q., & Yu, Y. (2007, August).
Co-clustering based classification for out-of-domain documents.
In Proceedings of the 13th ACM SIGKDD international conference on Knowledge discovery and data mining (pp. 210-219). ACM.
"""
picklepath = join(dataset_home, 'sraa.pkl')
if dopickle and exists(picklepath):
print('...loading pickle from {}'.format(picklepath))
return pickle.load(open(picklepath, 'rb'))
dataset_path = join(dataset_home, 'sraa.tar.gz')
if not exists(dataset_path):
create_if_not_exist(dataset_home)
print("downloading Reuters dataset (once and for all) into %s" %
dataset_path)
download_file(DOWNLOAD_URL, dataset_path)
print("untarring dataset...")
tarfile.open(dataset_path, 'r:gz').extractall(dataset_home)
sraa = load_files(join(dataset_home, 'sraa'), encoding='latin1')
remove = ('headers', 'footers') #, 'quotes')
if 'headers' in remove:
sraa.data = [strip_newsgroup_header(text) for text in sraa.data]
if 'footers' in remove:
sraa.data = [strip_newsgroup_footer(text) for text in sraa.data]
if 'quotes' in remove:
sraa.data = [strip_newsgroup_quoting(text) for text in sraa.data]
sraa = _index_by_label(sraa, min_words=10)
if dopickle:
print('...pickling the dataset into {} to speed-up next calls'.format(
picklepath))
pickle.dump(sraa, open(picklepath, 'wb'), pickle.HIGHEST_PROTOCOL)
return sraa
def preprocess_input(text):
text = strip_newsgroup_header(text)
text = strip_newsgroup_quoting(text)
text = strip_newsgroup_footer(text)
return text
import seaborn as sns
from sklearn.datasets import fetch_20newsgroups
from sklearn.datasets.twenty_newsgroups import strip_newsgroup_header, strip_newsgroup_quoting, strip_newsgroup_footer
newsgroups_train = fetch_20newsgroups(subset='train')['data']
newsgroups_test = fetch_20newsgroups(subset='test')['data']
NUM_TOPICS = 30
NUM_NEIGHBORS = 10
BUCKET = 'sagemaker-sbn'
PREFIX = '20newsgroups'
# In[ ]:
for i in range(len(newsgroups_train)):
newsgroups_train[i] = strip_newsgroup_header(newsgroups_train[i])
newsgroups_train[i] = strip_newsgroup_quoting(newsgroups_train[i])
newsgroups_train[i] = strip_newsgroup_footer(newsgroups_train[i])
# In[ ]:
newsgroups_train[1]
# In[ ]:
get_ipython().system('pip install nltk')
import nltk
nltk.download('punkt')
nltk.download('wordnet')
from nltk import word_tokenize
from nltk.stem import WordNetLemmatizer
import re
token_pattern = re.compile(r"(?u)\b\w\w+\b")