def combine_stopwords(dataframe_in, stopword_dict):
"""
Please use the stopwords() function and input that into the stopword_dict parameter.
returns filtered tokens.
:param dataframe_in:
:param stopword_dict:
:return:
"""
nlp = spacy.load("en_core_web_lg")
# Tokenizer
tokenizer = Tokenizer(nlp.vocab)
tokens = []
for doc in tokenizer.pipe(dataframe_in, batch_size=500):
doc_tokens = []
for token in doc:
if token.text.lower() not in stopword_dict:
doc_tokens.append(token.text.lower())
tokens.append(doc_tokens)
return tokens
def predict(): #define a prediction function
body = body.str.replace
tokenizer = Tokenizer(nlp.vocab)
tokens = []
""" Make them tokens """
#stop words
STOP_WORDS = nlp.Defaults.stop_words.union(
['', ' ', '-', 'reddit', 'post'])
tokens = []
for doc in tokenizer.pipe(df['combo'], batch_size=500):
doc_tokens = []
for token in doc:
if ((token.text.lower() not in STOP_WORDS)
and (token.is_stop == False)
and (token.is_punct == False)
and (token.pos_ != 'PRON')):
doc_tokens.append(token.lemma_.lower())
tokens.append(' '.join(doc_tokens))
df['tokens'] = tokens
tfidf = TfidfVectorizer(min_df=0.025, max_df=.98, ngram_range=(1, 2))
vec_text = tfidf.transform(user_input)
output = model.predict(vec_text.todense())
# give output to sender.
return jsonify({"response": output})
def _tokenizer(df):
nlp = English()
tokenizer = Tokenizer(nlp.vocab)
for doc in tokenizer.pipe(df.values.tolist(), batch_size=50):
for token in doc:
yield token
def tokenize_data(input_data):
nlp = spacy.load("en")
tokenizer = Tokenizer(nlp.vocab)
string_data = [str(data) for data in input_data]
tokenized_data = [[str(w) for w in doc]
for doc in tokenizer.pipe(string_data, batch_size=50)]
return tokenized_data
def train_tokenizer_mldoc(train_size=1000, datapath='../data/', savepath=None):
# Training a tokenizer for MLDoc dataset: create a vocabulary for each language in MLDoc
languages = [
"english", "german", "spanish", "french", "italian", "russian",
"chinese", "japanese"
]
mldoc_folder = os.path.join(datapath, 'mldoc')
savepath = os.path.join(mldoc_folder,
'vocab') if savepath is None else savepath
for lang in languages:
print("\n\n\t\t *** Training tokenizer for {} ***".format(lang))
train_f = os.path.join(mldoc_folder,
lang + '.train.{}'.format(train_size))
print('loading data')
train = pd.read_csv(train_f,
delimiter='\t',
header=None,
names=["label", "text"])
print('data: {}'.format(train.shape))
tokenizer = Tokenizer(language=lang,
train_list=train['text'].tolist(),
ngram_range=(1, 1),
min_freq=1,
max_freq_perc=1.0,
vocab_savefolder=savepath)
print("creating new tokenizer")
tokenizer2 = Tokenizer(language=lang, vocab_loadfolder=savepath)
print('loaded vocab: {}'.format(len(tokenizer2.word2ind)))
def train_tokenizer_twitter_sent(datapath='../data/', savepath=None):
languages = [
'arabic', 'bulgarian', 'german', 'english', 'spanish', 'persian',
'croatian', 'hungarian', 'polish', 'portuguese', 'russian', 'slovak',
'slovenian', 'swedish', 'uyghur', 'chinese'
]
savepath = os.path.join(
datapath, 'twitter_sent/vocab') if savepath is None else savepath
for lang in languages:
print("\n\n\t\t *** Training tokenizer for {} ***".format(lang))
train = load_df_twitter_sent(method='train',
language=lang,
print_fn=print)
print('data: {}'.format(train.shape))
# NOTE: We use tokenizer_method = 'clean' because data are already tokenized
tokenizer = Tokenizer(language=lang,
train_list=train['text'].tolist(),
tokenizer_method='clean',
remove_stopwords=False,
ngram_range=(1, 1),
min_freq=5,
max_freq_perc=1.0,
vocab_savefolder=savepath)
print("creating new tokenizer")
tokenizer2 = Tokenizer(language=lang,
vocab_loadfolder=savepath,
tokenizer_method='clean',
remove_stopwords=False)
print('loaded vocab: {}'.format(len(tokenizer2.word2ind)))
return
def test_tokenizer_add_special_case_tag(text, tokens):
vocab = Vocab()
tokenizer = Tokenizer(vocab, {}, None, None, None)
tokenizer.add_special_case(text, tokens)
doc = tokenizer(text)
assert doc[0].text == tokens[0]["orth"]
assert doc[0].norm_ == tokens[0]["norm"]
assert doc[1].text == tokens[1]["orth"]
def __init__(self,
rootDir='.cache',
vectorPath='vectors',
tokenizerPath='tokenizer'):
self.vectorPath = Path.cwd() / rootDir / vectorPath
self.tokenizerPath = Path.cwd() / rootDir / tokenizerPath
self.tokenizer = Tokenizer(Vocab())
self.vectors = Vectors(shape=(41299, 300))
def test_tokenizer_add_special_case_tag(text, tokens):
vocab = Vocab(tag_map={"NN": {"pos": "NOUN"}})
tokenizer = Tokenizer(vocab, {}, None, None, None)
tokenizer.add_special_case(text, tokens)
doc = tokenizer(text)
assert doc[0].text == tokens[0]["orth"]
assert doc[0].tag_ == tokens[0]["tag"]
assert doc[0].pos_ == "NOUN"
assert doc[1].text == tokens[1]["orth"]
def test_tokenizer_flush_cache(en_vocab):
suffix_re = re.compile(r"[\.]$")
tokenizer = Tokenizer(
en_vocab,
suffix_search=suffix_re.search,
)
assert [t.text for t in tokenizer("a.")] == ["a", "."]
tokenizer.suffix_search = None
assert [t.text for t in tokenizer("a.")] == ["a."]
def transform(self, data):
tokenizer = Tokenizer(nlp.vocab)
return np.array([
np.mean([
self.model[w.text.lower()] * self.word2weight[w.text.lower()]
for w in words if w.text.lower() in self.model
] or [np.zeros(self.dim)],
axis=0) for words in tokenizer.pipe(data)
])
def test_tokenizer_add_special_case_tag(text, tokens):
vocab = Vocab(tag_map={"NN": {"pos": "NOUN"}})
tokenizer = Tokenizer(vocab, {}, None, None, None)
tokenizer.add_special_case(text, tokens)
doc = tokenizer(text)
assert doc[0].text == tokens[0]["orth"]
assert doc[0].tag_ == tokens[0]["tag"]
assert doc[0].pos_ == "NOUN"
assert doc[1].text == tokens[1]["orth"]
def get_lemmas(text):
# nlp = spacy.load("en_core_web_sm-2.2.5", path="airbnb_api/")
nlp = spacy.load("en_core_web_sm-2.2.5", path="./")
# nlp = spacy.load("en_core_web_sm")
# nlp = en_core_web_sm.load()
tokenizer = Tokenizer(nlp.vocab)
STOP_WORDS = nlp.Defaults.stop_words.union([
' ', 'und', '-', 'die', 'der', 'berlin', 'ein', 'das', 'mit', 'ist',
'im', 'zu', 'eine', 'es', 'für'
'berlin.', 'zum', 'sind', 'für', 'Berlin.', '-pron-', 's', 'u', '',
"'", ' ', '-PRON-'
])
lemmas = []
doc = nlp(text)
for token in doc:
lemmas.append(token.lemma_)
lemma_summary = []
working_set = ""
for lemma in lemmas:
working_set += lemma + ' '
lemma_summary.append(working_set)
description = [lemma_summary[0]]
tokens = []
for doc in tokenizer.pipe(description, batch_size=500):
doc_tokens = []
for token in doc:
if ((token.is_stop == False) and
(token.is_punct == False)) and (token.pos_ != 'PRON'):
if token.text.lower() not in STOP_WORDS:
doc_tokens.append(token.text.lower())
tokens.append(doc_tokens)
token_summary = []
for set_of_tokens in tokens:
working_set = ""
for variable in set_of_tokens:
working_set += variable + ' '
token_summary.append(working_set)
return token_summary[0]
def test_tokenizer_flush_specials(en_vocab):
suffix_re = re.compile(r"[\.]$")
rules = {"a a": [{"ORTH": "a a"}]}
tokenizer1 = Tokenizer(
en_vocab,
suffix_search=suffix_re.search,
rules=rules,
)
assert [t.text for t in tokenizer1("a a.")] == ["a a", "."]
tokenizer1.rules = {}
assert [t.text for t in tokenizer1("a a.")] == ["a", "a", "."]
def __init__(self, whitespace_tokenizer_for_coref=True, whitespace_tokenizer_for_tokenizer=False):
# for tokenization
self.tokenize_nlp = spacy.load('en')
if whitespace_tokenizer_for_tokenizer:
self.tokenize_nlp.tokenizer = Tokenizer(self.tokenize_nlp.vocab)
# for coreference resolution
self.whitespace_tokenizer_for_coref = whitespace_tokenizer_for_coref
self.coref_nlp = spacy.load('en')
if whitespace_tokenizer_for_coref:
self.coref_nlp.tokenizer = Tokenizer(self.coref_nlp.vocab)
def test_tokenizer_initial_special_case_explain(en_vocab):
tokenizer = Tokenizer(
en_vocab,
token_match=re.compile("^id$").match,
rules={
"id": [{"ORTH": "i"}, {"ORTH": "d"}],
},
)
tokens = [t.text for t in tokenizer("id")]
explain_tokens = [t[1] for t in tokenizer.explain("id")]
assert tokens == explain_tokens
def test_tokenizer_explain_special_matcher(en_vocab):
suffix_re = re.compile(r"[\.]$")
infix_re = re.compile(r"[/]")
rules = {"a.": [{"ORTH": "a."}]}
tokenizer = Tokenizer(
en_vocab,
rules=rules,
suffix_search=suffix_re.search,
infix_finditer=infix_re.finditer,
)
tokens = [t.text for t in tokenizer("a/a.")]
explain_tokens = [t[1] for t in tokenizer.explain("a/a.")]
assert tokens == explain_tokens
def test_spacy_tokenizer_pipe(nlp):
tokenizer = Tokenizer(nlp.vocab)
token_sets = []
for doc in tokenizer.pipe(DOCUMENTS, batch_size=2):
doc_tokens = [token.text for token in doc]
token_sets.append(doc_tokens)
assert token_sets == [['all', 'the', 'kings', 'men'],
['ate', 'all', 'the', 'kings', 'hens'],
[
'until', 'they', 'all', 'got', 'tired', 'and',
'went', 'to', 'sleep', 'zzz'
]]
def test_tokenizer_infix_prefix(en_vocab):
# the prefix and suffix matches overlap in the suffix lookbehind
infixes = ["±"]
suffixes = ["%"]
infix_re = compile_infix_regex(infixes)
suffix_re = compile_suffix_regex(suffixes)
tokenizer = Tokenizer(
en_vocab,
infix_finditer=infix_re.finditer,
suffix_search=suffix_re.search,
)
tokens = [t.text for t in tokenizer("±10%")]
assert tokens == ["±10", "%"]
explain_tokens = [t[1] for t in tokenizer.explain("±10%")]
assert tokens == explain_tokens
def test_tokenizer_prefix_suffix_overlap_lookbehind(en_vocab):
# the prefix and suffix matches overlap in the suffix lookbehind
prefixes = ["a(?=.)"]
suffixes = [r"(?<=\w)\.", r"(?<=a)\d+\."]
prefix_re = compile_prefix_regex(prefixes)
suffix_re = compile_suffix_regex(suffixes)
tokenizer = Tokenizer(
en_vocab,
prefix_search=prefix_re.search,
suffix_search=suffix_re.search,
)
tokens = [t.text for t in tokenizer("a10.")]
assert tokens == ["a", "10", "."]
explain_tokens = [t[1] for t in tokenizer.explain("a10.")]
assert tokens == explain_tokens
def tokenize(df_in):
"""
Tokenize by inputting a dataframe. Outputs a tokenized list.
:param df_in:
:return:
"""
nlp = spacy.load("en_core_web_lg")
# Tokenizer
tokenizer = Tokenizer(nlp.vocab)
tokens = []
for doc in tokenizer.pipe(df_in, batch_size=500):
doc_tokens = [token.text for token in doc]
tokens.append(doc_tokens)
return tokens
def __init__(self):
""" coppied from notebook at app/ml/Build_week_IsaacGrove.ipynb
"""
self.PICKLE_PATH = path.join(path.dirname(__file__), '..', 'pickles',
'')
# for now i'm loading data from a static link, will try to pull live data
# in future iters
leafly = pd.read_csv(
'https://raw.githubusercontent.com/Build-Week-Med-Cabinet-6/DS/mark-dev/data/cannabis.csv'
)
# Set up spacy tokenizer
nlp = English()
tokenizer = Tokenizer(nlp.vocab)
# work around for pickle
self.nlp = nlp
# clean some missing info
leafly.replace('None', np.NaN, inplace=True)
leafly = leafly.dropna()
# Make tokens out of descriptions
tokens = []
for desc in tokenizer.pipe(leafly['Description'], batch_size=500):
desc_tokens = [token.text for token in desc]
tokens.append(desc_tokens)
leafly['tokens'] = tokens
leafly['tokens'].head()
# Instantiate vectorizer object
tfidf = TfidfVectorizer(ngram_range=(1, 2),
max_df=.7,
min_df=.001,
tokenizer=self.tokenize)
# Create a vocabulary and get word counts per listing
dtm = tfidf.fit_transform(leafly['Description'])
# Get feature names to use a dataframe column headers
dtm = pd.DataFrame(dtm.todense(), columns=tfidf.get_feature_names())
# Fit on dtm
nn = NearestNeighbors(n_neighbors=20, algorithm='kd_tree')
nn.fit(dtm)
self.model = nn
self.transform = tfidf
return
def __init__(self, language):
self.nlp = load_spacy(language)
custom_infixes = ['/']
all_prefixes_re = spacy.util.compile_prefix_regex(tuple(list(self.nlp.Defaults.prefixes) + custom_infixes))
infix_re = spacy.util.compile_infix_regex(tuple(list(self.nlp.Defaults.infixes) + custom_infixes))
suffix_re = spacy.util.compile_suffix_regex(tuple(list(self.nlp.Defaults.suffixes) + custom_infixes))
self.nlp.tokenizer = Tokenizer(self.nlp.vocab, self.nlp.Defaults.tokenizer_exceptions,
prefix_search = all_prefixes_re.search,
infix_finditer = infix_re.finditer,
suffix_search = suffix_re.search,
token_match=None)
self.matcher = Matcher(self.nlp.vocab)
self.matcher.add('morphology', None, [
{"TEXT": {"REGEX" : r'^\d\d\d\d$'}},
{"TEXT": {"REGEX" : r'\s'}, "OP" : "*"},
{"TEXT": '/'},
{"TEXT": {"REGEX" : r'\s'}, "OP" : "*"},
{"TEXT": {"REGEX" : r'\d'}},
])
def clean_mag_data(dataframe, save_path):
samples = []
# prepare tokenization functions
nlp = spacy.load("en_core_web_lg")
tokenizer = Tokenizer(nlp.vocab)
#take samples with at least 10 words in citation context
for index, row in dataframe.iterrows():
context = row['context']
text = re.sub("[" + re.escape(string.punctuation) + "]", " ", context)
text = [
token.lemma_ for token in tokenizer(text) if not token.like_num
]
text = [token for token in text if token.strip()]
if (len(text) < MIN_CONTEXT_LENGTH):
continue
# generate sample in correct format
#"paper_id": row['paperid'],
sample = {
"context": context,
"authors_citing": row['citingauthors'],
"title_cited": row['citedtitle'],
"authors_cited": row['citedauthors']
}
samples.append(pd.DataFrame(sample, index=[0]))
logger.info("mag samples ready to load to file...")
dataset = pd.concat(samples, axis=0)
dataset.to_csv(save_path, compression=None, index=False, index_label=False)
def __init__(self, rollout_num, vocab):
#self.new_net = copy.deepcopy(net)
self.vocab = vocab
self.tokenizer = Tokenizer(
Vocab(strings=list(vocab.labelToIdx.keys())))
self.rollout_num = rollout_num
self.parser = StanfordParser(annots='tokenize')
def transform(self, data):
tokenizer = Tokenizer(nlp.vocab)
return np.array(
[
np.mean(
[
self.model[w.text.lower()]
for w in words
if w.text.lower() in self.model
]
or [np.zeros(self.dim)],
axis=0,
)
for words in tokenizer.pipe(data)
]
)
def search_func(user_input, num_results=10):
"""
Flexible function that searches for cannabis strains.
### Request Body
- user_input str
- num_results int: default 10
### Response
- `strain_recommendation`: dictionary of strain recommendations
"""
user_input = [user_input]
nlp = English()
tokenizer = Tokenizer(nlp.vocab)
tf = TfidfVectorizer(stop_words='english')
dtm = tf.fit_transform(df['search'])
dtm = pd.DataFrame(dtm.todense(), columns=tf.get_feature_names())
nr = num_results
nn = NearestNeighbors(n_neighbors=nr, algorithm='ball_tree')
nn.fit(dtm)
dtf = tf.transform(user_input)
_, output = nn.kneighbors(dtf.todense())
recommendations = []
for n in output:
for row in n:
recommendations.append(row)
result = []
for i in recommendations:
data = (df.loc[i, :])
result.append(data)
return {'strain_recommendations': result}
def load_data(data_path: str,
tokenize: bool = False,
tokenizer_type: str = "just_spaces") -> List[str]:
if tokenizer_type == "just_spaces":
tokenizer = SpacyWordSplitter()
elif tokenizer_type == "spacy":
nlp = spacy.load('en')
tokenizer = Tokenizer(nlp.vocab)
tokenized_examples = []
with tqdm(open(data_path, "r"), desc=f"loading {data_path}") as f:
for line in f:
if data_path.endswith(".jsonl") or data_path.endswith(".json"):
example = json.loads(line)
else:
example = {"text": line}
if tokenize:
if tokenizer_type == 'just_spaces':
tokens = list(
map(str, tokenizer.split_words(example['text'])))
elif tokenizer_type == 'spacy':
tokens = list(map(str, tokenizer(example['text'])))
text = ' '.join(tokens)
else:
text = example['text']
tokenized_examples.append(text)
return tokenized_examples
def _create_tokenizer(nlp):
infix_re = spacy.util.compile_infix_regex(
TOKENIZER_INFIXES + [ # u'\w*[,-.–_—:;\(\)\[\]\{\}/]{1,3}\S\w*',
# r'\w*[,\-.\-_:;\(\)\[\]\{\}\/]{1,3}\S\w*',
# r'((?P<start_with_non_whitespace_and_one_or_more_punctation>\b\S+|[,.-_:;\(\)\[\]\{\}/\+])(
# ?P<has_1_or_more_punctation>[,.-_:;\(\)\[\]\{\}/\+])+(
# ?P<ends_with_non_whitespace_or_non_terminating_punctation>\S+\b[,.-_:;\(\)\[\]\{\}/\+]|[,.-_:;\(\)\[
# \]\{\}/\+|\-]|\S+\b))',
# r'\w*\S-\S*\w',
# u'\w*\S–\S*\w',
# u'\w*\S—\S*\w',
# u'\w*[,-.–_—:;\(\)\[\]\{\}/]{1,3}\S\w*'
ur'(?P<start_with_non_whitespace_and_one_or_more_punctation>\b\S*|[,.-_-:–;—\(\[\{/\+]?)('
ur'?P<has_1_or_more_punctation>[,.-_-:–;—\(\)\[\]\{\}/\+])+('
ur'?P<ends_with_non_whitespace_or_non_terminating_punctation>\S+\b[,.-_-:–;—\)\]\}/\+]|[,'
ur'.-_-:–;—\)\]\}/\+}]|\S+\b)'
])
# TODO: prefix and suffix raise TypeError: '_regex.Pattern' object is not callable
# prefix_boundaries_to_keep = ur'\) \] \} \> , . - _ - : – ; — \+ -'.split()
# suffix_boundaries_to_keep = ur'\( \[ \{ \< , . - _ - : – ; — \+ -'.split()
# prefixe_re = spacy.util.compile_prefix_regex([i for i in TOKENIZER_PREFIXES if i not in
# prefix_boundaries_to_keep])
# suffixe_re = spacy.util.compile_suffix_regex([i for i in TOKENIZER_SUFFIXES if i not in
# suffix_boundaries_to_keep])
#
# return Tokenizer(nlp.vocab, {}, prefixe_re.search, suffixe_re.search,
# infix_re.finditer)
return Tokenizer(nlp.vocab, {}, nlp.tokenizer.prefix_search,
nlp.tokenizer.suffix_search, infix_re.finditer)
def get_spacy_tokens(self, spacy_sentence, target_word, spacy_model):
"""
A function to locate the target phrase spacy tokens in a spacy doc of
a whole sentence.
Args:
spacy_sentence: spacy doc for the context sentence
spacy_target_word: spacy doc for the target word/phrase only
Returns:
spacy_token_list: a list of the spacy tokens for the target phrase,
using the information from the context sentence.
"""
# Create the tokeniser
tokenizer = Tokenizer(spacy_model.vocab)
spacy_token_list = []
for target in tokenizer(target_word):
for wd in spacy_sentence:
if target.text == wd.text:
spacy_token_list.append(wd)
break
return spacy_token_list
def test_create(self):
vocab = Vocab()
dummy_re = re.compile(r'sklfb;s')
tokenizer = Tokenizer(vocab, {}, dummy_re, dummy_re, dummy_re)
doc = tokenizer(u'I am a document.')
self.assertEqual(len(doc), 4)
def custom_tokenizer(self):
""" Function that defines a tokenizer in order to be used
Parameters
-----------
nlp: spacy loaded object
return: prepared tokenizer
"""
infixes = (
LIST_ELLIPSES + LIST_ICONS + [
r"(?<=[0-9])[+\-\*^](?=[0-9-])",
r"(?<=[{al}{q}])\.(?=[{au}{q}])".format(
al=ALPHA_LOWER, au=ALPHA_UPPER, q=CONCAT_QUOTES),
r"(?<=[{a}]),(?=[{a}])".format(a=ALPHA),
#r"(?<=[{a}])(?:{h})(?=[{a}])".format(a=ALPHA, h=HYPHENS),
r"(?<=[{a}0-9])[:<>=/](?=[{a}])".format(a=ALPHA),
])
infix_re = compile_infix_regex(infixes)
return Tokenizer(self.nlp.vocab,
prefix_search=self.nlp.tokenizer.prefix_search,
suffix_search=self.nlp.tokenizer.suffix_search,
infix_finditer=infix_re.finditer,
token_match=self.nlp.tokenizer.token_match,
rules=self.nlp.Defaults.tokenizer_exceptions)
def main(output_dir):
ensure_dir(output_dir)
ensure_dir(output_dir, "pos")
ensure_dir(output_dir, "vocab")
vocab = Vocab(tag_map=TAG_MAP)
tokenizer = Tokenizer(vocab, {}, None, None, None)
# The default_templates argument is where features are specified. See
# spacy/tagger.pyx for the defaults.
tagger = Tagger.blank(vocab, Tagger.default_templates())
for i in range(5):
for words, tags in DATA:
tokens = tokenizer.tokens_from_list(words)
tagger.train(tokens, tags)
random.shuffle(DATA)
tagger.model.end_training(path.join(output_dir, 'pos', 'model'))
vocab.strings.dump(path.join(output_dir, 'vocab', 'strings.txt'))
def count_freqs(input_loc, output_loc):
print(output_loc)
vocab = English.default_vocab(get_lex_attr=None)
tokenizer = Tokenizer.from_dir(vocab,
path.join(English.default_data_dir(), 'tokenizer'))
counts = PreshCounter()
for json_comment in iter_comments(input_loc):
doc = tokenizer(json_comment['body'])
doc.count_by(ORTH, counts=counts)
with io.open(output_loc, 'w', 'utf8') as file_:
for orth, freq in counts:
string = tokenizer.vocab.strings[orth]
if not string.isspace():
file_.write('%d\t%s\n' % (freq, string))
def test_serialize_custom_tokenizer(en_vocab, en_tokenizer):
"""Test that custom tokenizer with not all functions defined can be
serialized and deserialized correctly (see #2494)."""
tokenizer = Tokenizer(en_vocab, suffix_search=en_tokenizer.suffix_search)
tokenizer_bytes = tokenizer.to_bytes()
Tokenizer(en_vocab).from_bytes(tokenizer_bytes)
def test_load(self):
data_dir = English.default_data_dir()
vocab = Vocab.from_dir(path.join(data_dir, 'vocab'))
tokenizer = Tokenizer.from_dir(vocab, path.join(data_dir, 'tokenizer'))
