def test_ngrams():
# bigrams
strings = ["this is my favorite", "book on my bookshelf"]
dstrings = nvstrings.to_device(strings)
expected = [
"this_is",
"is_my",
"my_favorite",
"favorite_book",
"book_on",
"on_my",
"my_bookshelf",
]
tokens = nvtext.tokenize(dstrings)
outcome = nvtext.ngrams(tokens, N=2, sep="_")
assert outcome.to_host() == expected
# trigrams
strings = ["this is my favorite", "book on my bookshelf"]
dstrings = nvstrings.to_device(strings)
expected = [
"this-is-my",
"is-my-favorite",
"my-favorite-book",
"favorite-book-on",
"book-on-my",
"on-my-bookshelf",
]
tokens = nvtext.tokenize(dstrings)
outcome = nvtext.ngrams(tokens, N=3, sep="-")
assert outcome.to_host() == expected
def test_ngrams():
# bigrams
strings = ['this is my favorite', 'book on my bookshelf']
dstrings = nvstrings.to_device(strings)
expected = [
'this_is',
'is_my',
'my_favorite',
'favorite_book',
'book_on',
'on_my',
'my_bookshelf'
]
tokens = nvtext.tokenize(dstrings)
outcome = nvtext.ngrams(tokens, N=2, sep='_')
assert outcome.to_host() == expected
# trigrams
strings = ['this is my favorite', 'book on my bookshelf']
dstrings = nvstrings.to_device(strings)
expected = [
'this-is-my',
'is-my-favorite',
'my-favorite-book',
'favorite-book-on',
'book-on-my',
'on-my-bookshelf'
]
tokens = nvtext.tokenize(dstrings)
outcome = nvtext.ngrams(tokens, N=3, sep='-')
assert outcome.to_host() == expected
def test_tokenize():
# default space delimiter
strs = nvstrings.to_device(
[
"the quick fox jumped over the lazy dog",
"the siamésé cat jumped under the sofa",
None,
"",
]
)
outcome = nvtext.tokenize(strs)
expected = [
"the",
"quick",
"fox",
"jumped",
"over",
"the",
"lazy",
"dog",
"the",
"siamésé",
"cat",
"jumped",
"under",
"the",
"sofa",
]
assert outcome.to_host() == expected
def get_word_count(str_col):
"""
returns the count of input strings
"""
## Tokenize: convierte oraciones en largas listas de palabras
df = cudf.DataFrame()
principales = []
for data in str_col:
principales.append(data)
# Tokenizando strings con nvstring y llevandolo al gpu con to_device
letras = nvstrings.to_device(principales)
df['string'] = nvtext.tokenize(letras)
# Se usa group by para realizar el conteo de palabras por columna
# Nativamente sera soportado pronto
# Mirar para errores: https://github.com/rapidsai/cudf/issues/1951
df['counts'] = np.dtype('int32').type(0)
res = df.groupby('string').count()
res = res.reset_index(drop=False).sort_values(by='counts', ascending=False)
return res.rename({'index':'string'})
