def test_prune(self):
t_docs = [['cat', 'cat dog', 'happy', 'dog', 'dog'],
['cat', 'cat dog', 'sad']]
expected_t_docs = [['cat dog', 'happy', 'dog', 'dog'],
['cat dog', 'sad']]
t_docs = util.prune(t_docs)
self.assertEqual(t_docs, expected_t_docs)
def tokenize(self, docs):
if self.lemmatize:
lem = WordNetLemmatizer()
pre_tdocs = RAKE().tokenize(docs)
tdocs = []
for i, tdoc in enumerate(pre_tdocs):
# Split phrase keywords into 1gram keywords,
# to check tokens against
kws_1g = [t.split(' ') for t in tdoc]
kws_1g = [kw for grp in kws_1g for kw in grp]
toks = spacy(docs[i], tag=True, parse=False, entity=False)
tagged = [(t.lower_.strip(), t.tag_) for t in toks]
toks = []
for tok, tag in tagged:
if tok in kws_1g:
wn_tag = penn_to_wordnet(tag)
if wn_tag is not None:
toks.append(lem.lemmatize(tok, wn_tag))
tdocs.append(toks)
tdocs = extract_phrases(tdocs, docs)
if prune:
return prune(tdocs)
return tdocs
def test_prune(self):
t_docs = [
['cat', 'cat dog', 'happy', 'dog', 'dog'],
['cat', 'cat dog', 'sad']
]
expected_t_docs = [
['cat dog', 'happy', 'dog', 'dog'],
['cat dog', 'sad']
]
t_docs = util.prune(t_docs)
self.assertEqual(t_docs, expected_t_docs)
def tokenize(self, docs):
tags = ['NN', 'NNS', 'NNP', 'NNPS']
keywords = []
for doc in docs:
toks = spacy(doc, tag=True, parse=False, entity=False)
tagged = [(t.lower_.strip(), t.tag_) for t in toks]
kws = [t for t, tag in tagged if tag in tags]
kws += extract_noun_phrases(tagged)
keywords.append(kws)
return prune(keywords)
def tokenize(self, docs):
"""
The first pass consists of converting documents
into "transactions" (sets of their tokens)
and the initial frequency/support filtering.
Then iterate until we close in on a final set.
`docs` can be any iterator or generator so long as it yields lists.
Each list represents a document (i.e. is a list of tokens).
For example, it can be a list of lists of nouns and noun phrases if trying
to identify aspects, where each list represents a sentence or document.
`min_sup` defines the minimum frequency (as a ratio over the total) necessary to
keep a candidate.
"""
if self.min_sup < 1 / len(docs):
raise Exception(
'`min_sup` must be greater than or equal to `1/len(docs)`.')
# First pass
candidates = set()
transactions = []
# Use nouns and noun phrases.
for doc in POSTokenizer().tokenize(docs):
transaction = set(doc)
candidates = candidates.union({(t, ) for t in transaction})
transactions.append(transaction)
freq_set = filter_support(candidates, transactions, self.min_sup)
# Iterate
k = 2
last_set = set()
while freq_set != set():
last_set = freq_set
cands = generate_candidates(freq_set, k)
freq_set = filter_support(cands, transactions, self.min_sup)
k += 1
# Map documents to their keywords.
keywords = flatten(last_set)
return prune([[kw for kw in keywords if kw in doc] for doc in docs])
def tokenize(self, docs):
"""
The first pass consists of converting documents
into "transactions" (sets of their tokens)
and the initial frequency/support filtering.
Then iterate until we close in on a final set.
`docs` can be any iterator or generator so long as it yields lists.
Each list represents a document (i.e. is a list of tokens).
For example, it can be a list of lists of nouns and noun phrases if trying
to identify aspects, where each list represents a sentence or document.
`min_sup` defines the minimum frequency (as a ratio over the total) necessary to
keep a candidate.
"""
if self.min_sup < 1/len(docs):
raise Exception('`min_sup` must be greater than or equal to `1/len(docs)`.')
# First pass
candidates = set()
transactions = []
# Use nouns and noun phrases.
for doc in POSTokenizer().tokenize(docs):
transaction = set(doc)
candidates = candidates.union({(t,) for t in transaction})
transactions.append(transaction)
freq_set = filter_support(candidates, transactions, self.min_sup)
# Iterate
k = 2
last_set = set()
while freq_set != set():
last_set = freq_set
cands = generate_candidates(freq_set, k)
freq_set = filter_support(cands, transactions, self.min_sup)
k += 1
# Map documents to their keywords.
keywords = flatten(last_set)
return prune([[kw for kw in keywords if kw in doc] for doc in docs])
