def get_negative_labeling_function(divisor: int) -> LabelingFunction:
"""Get LabelingFunction that abstains unless x0 is divisible by divisor."""
def f(x):
return 0 if x.x0 % divisor == 0 and x.x1 <= x.x2 + 0.25 else -1
return LabelingFunction(f"lf_neg_{divisor}", f)
def make_keyword_lf_NL(keywords, label=NL):
return LabelingFunction(
name=f"keyword_{keywords[0]}_NL",
f=keyword_lookup,
resources=dict(keywords=keywords, label=label),
# pre=[spacy_preproc]
)
def make_thresold_lf(thresh, col_name, next_threshold=sys.maxsize):
return LabelingFunction(
name="more_than_%s_%s" % (thresh, col_name),
f=more_than_treshold,
resources=dict(thresh=thresh,
col_name=col_name,
next_threshold=next_threshold),
)
def wrapper(func):
# set up kwargs for Snorkel's LF
# a default name that can be overridden
snorkel_kwargs = {"name": func.__name__}
snorkel_kwargs.update(kwargs)
# return value of hover's decorator
lf = SnorkelLF(f=func, **snorkel_kwargs)
# additional attributes
lf.uuid = uuid.uuid1()
lf.targets = targets[:]
# link a snorkel-style labeling function if applicable
if label_encoder:
lf.label_encoder = label_encoder
def snorkel_style_func(x):
return lf.label_encoder[func(x)]
lf.snorkel = snorkel_lf(**kwargs)(snorkel_style_func)
else:
lf.label_encoder = None
lf.snorkel = None
return lf
def make_worker_lf(worker_id,
x_id_field: str = "tweet_id") -> LabelingFunction:
def worker_lf(x, worker_dict):
return worker_dict.get(x[x_id_field], ABSTAIN)
worker_dict = worker_dicts[worker_id]
name = f"worker_{worker_id}"
return LabelingFunction(name,
f=worker_lf,
resources={"worker_dict": worker_dict})
def createAnalysis(final_df, category_names):
L_final = []
for name in category_names:
category_if = [-1 if i == 0 else i for i in final_df[name].tolist()]
L_final.append(category_if)
L_train = [list(x) for x in list(zip(*L_final))]
lfs = [LabelingFunction(name=name, f=None) for name in category_names]
return LFAnalysis(L=np.array(L_train), lfs=lfs).lf_summary()
def make_keyword_lf(keyword, label, neg_label, context_len, with_period):
return LabelingFunction(
name="pattern_%s_%s%s" % (keyword, "context:%d" % context_len,
"_period" if with_period else ""),
f=pattern_match,
resources=dict(keyword=keyword,
label=label,
neg_label=neg_label,
with_period=with_period,
context_len=context_len),
)
def make_keyword_lf(keywords, label=1):
def keyword_lookup(x, keywords, label):
if any(word in x.text.lower() for word in keywords):
return label
return -1
return LabelingFunction(
name=f"keyword_{keywords[0]}",
f=keyword_lookup,
resources=dict(keywords=keywords, label=label),
)
def make_annotator_lf(worker_index, num_annotators = sys.maxsize):
worker_index_old = worker_index
worker_index = worker_index % num_annotators
reader = csv.reader(open("/home/tigunova/PycharmProjects/snorkel_labels/data/hobby/labeling_lf/%d.csv" % (worker_index)))
next(reader)
worker_dict = dict((x[-1], POS if x[1] == "checked" else NEG) for x in reader)
return LabelingFunction(
name="worker_%d" % (worker_index_old),
f=worker_lf,
resources=dict(worker_dict=worker_dict),
)
def make_keyword_lf(keywords: List[str], label: str, field: str = "text"):
def keyword_lookup(x, keywords, label):
if any(word in x[field].lower() for word in keywords):
return label
return ABSTAIN
return LabelingFunction(
name=f"keyword_{keywords[0]}",
f=keyword_lookup,
resources=dict(keywords=keywords, label=label),
)
def make_annotator_lf(worker_index):
reader = csv.reader(
open(
"/home/tigunova/PycharmProjects/snorkel_labels/data/profession/labeling_lf/%d.csv"
% (worker_index)))
next(reader)
worker_dict = dict(
(x[-1], POS if x[1] == "checked" else ABSTAIN) for x in reader)
return LabelingFunction(
name="worker_%d" % (worker_index),
f=worker_lf,
resources=dict(worker_dict=worker_dict),
)
def make_keyword_lf(keywords: list, label: int = RELEVANT):
"""Generate labeling functions from keywords related to the application area
:param keywords: A list of keywords related to the application area
:type keywords: list
:param label: The label that should be assigned to each labeling function, defaults to RELEVANT
:type label: int, optional
:return: Returns a labeling function which implements the `keyword_lookup` function
:rtype: LabelingFunction
"""
return LabelingFunction(name=f"keyword_{keywords[0]}",
f=keyword_lookup,
resources=dict(keywords=keywords, label=label))
def make_keyword_lf(self, keywords: list, label: int, lf_name: str = None):
"""Generate a labeling function from a keyword
:param keywords: A list of keywords which will be used to generate the labeling functions
:type keywords: list
:param label: The label to assign to the labeling function
:type label: int
:param lf_name: A unique name for the labling function
:type lf_name: str, optional
:return: returns a labeling function which implements `keyword_lookup`
:rtype: LabelingFunction
"""
labeling_function_name = f"keyword_{re.sub(' ', '_', keywords[0].strip())}"
return LabelingFunction(name=labeling_function_name,
f=self.keyword_lookup,
resources=dict(keywords=keywords, label=label))
def test_lf_summary(self) -> None:
df = self.lfa.lf_summary(self.Y, est_weights=None)
df_expected = pd.DataFrame(
{
"Polarity": [[1, 2], [], [0, 2], [2], [0, 1], [0]],
"Coverage": [3 / 6, 0, 3 / 6, 2 / 6, 2 / 6, 4 / 6],
"Overlaps": [3 / 6, 0, 3 / 6, 1 / 6, 2 / 6, 4 / 6],
"Conflicts": [3 / 6, 0, 2 / 6, 1 / 6, 2 / 6, 3 / 6],
"Correct": [1, 0, 1, 1, 1, 2],
"Incorrect": [2, 0, 2, 1, 1, 2],
"Emp. Acc.": [1 / 3, 0, 1 / 3, 1 / 2, 1 / 2, 2 / 4],
}
)
pd.testing.assert_frame_equal(df.round(6), df_expected.round(6))
df = self.lfa.lf_summary(Y=None, est_weights=None)
df_expected = pd.DataFrame(
{
"Polarity": [[1, 2], [], [0, 2], [2], [0, 1], [0]],
"Coverage": [3 / 6, 0, 3 / 6, 2 / 6, 2 / 6, 4 / 6],
"Overlaps": [3 / 6, 0, 3 / 6, 1 / 6, 2 / 6, 4 / 6],
"Conflicts": [3 / 6, 0, 2 / 6, 1 / 6, 2 / 6, 3 / 6],
}
)
pd.testing.assert_frame_equal(df.round(6), df_expected.round(6))
est_weights = [1, 0, 1, 1, 1, 0.5]
names = list("abcdef")
lfs = [LabelingFunction(s, f) for s in names]
lfa = LFAnalysis(np.array(L), lfs)
df = lfa.lf_summary(self.Y, est_weights=est_weights)
df_expected = pd.DataFrame(
{
"j": [0, 1, 2, 3, 4, 5],
"Polarity": [[1, 2], [], [0, 2], [2], [0, 1], [0]],
"Coverage": [3 / 6, 0, 3 / 6, 2 / 6, 2 / 6, 4 / 6],
"Overlaps": [3 / 6, 0, 3 / 6, 1 / 6, 2 / 6, 4 / 6],
"Conflicts": [3 / 6, 0, 2 / 6, 1 / 6, 2 / 6, 3 / 6],
"Correct": [1, 0, 1, 1, 1, 2],
"Incorrect": [2, 0, 2, 1, 1, 2],
"Emp. Acc.": [1 / 3, 0, 1 / 3, 1 / 2, 1 / 2, 2 / 4],
"Learned Weight": [1, 0, 1, 1, 1, 0.5],
}
).set_index(pd.Index(names))
pd.testing.assert_frame_equal(df.round(6), df_expected.round(6))
def get_lfs(keywords):
lfs = []
idx_label_map = {}
for i, (label, kws) in enumerate(keywords.items()):
idx_label_map[i] = label
for kw in kws:
parsed_kw = parse_kw(kw)
if parsed_kw['first'] == 'B':
name = parsed_kw['text'] + f'...[{label}]'
elif parsed_kw['first'] == 'I':
name = f'[{parsed_kw["text"]}...{label}]'
elif parsed_kw['first'] == 'A':
name = f'[{label}]...{parsed_kw["text"]}'
lfs.append(
LabelingFunction(name=name,
f=sent_context_lf,
resources=dict(label=i, parsed_kw=parsed_kw)))
return lfs, idx_label_map
def pos(sample):
return 1 if re.search(PWORDS, str(sample)) else -1
NWORDS = r"\b(hell yeah|bribery|not happy|less moral|impeach trump|impeach our president)"
def neg(sample):
return 0 if re.search(NWORDS, str(sample)) else -1
# In[8]:
positive = LabelingFunction(f"positive", f=pos)
# In[9]:
negative = LabelingFunction(f"negative", f=neg)
# In[10]:
df_train = dt[0:450000]
# In[11]:
# Create the labeling functions using the textblob sentiment analyzer
@preprocessor(memoize=True)
def textblob_polarity(x):
def make_keyword_lf(keywords, label=SPAM):
return LabelingFunction(
name=f"keyword_{keywords[0]}",
f=keyword_lookup,
resources=dict(keywords=keywords, label=label),
)
def test_wrong_number_of_lfs(self) -> None:
with self.assertRaisesRegex(ValueError, "Number of LFs"):
LFAnalysis(np.array(L), [LabelingFunction(s, f) for s in "ab"])
def make_keyword_lf(lf_name, keywords, label=IRRELEVANT):
return LabelingFunction(
name=lf_name,
f=regex_keyword_lookup,
resources=dict(keywords=keywords, label=label),
)
def make_abstract_lf(keywords, name, label=None):
return LabelingFunction(
name=f"abstract_{name}",
f=abstract_lookup,
resources=dict(keywords=keywords, label=label),
)
def make_keyword_lf(keywords, virus, name, label=None):
return LabelingFunction(
name=f"keyword_{name}",
f=keyword_lookup,
resources=dict(keywords=keywords, virus=virus, label=label),
)
def make_lfs_list(post_hoc_callables, GI_callables, rule_out_callables,
lab_callables):
lfs = []
for f in post_hoc_callables:
lfs.append(f)
for f in rule_out_callables:
lfs.append(f)
for f in GI_callables:
lfs.append(f)
for f in lab_callables:
lfs.append(f)
# """
# Returns 1 if any rule out condition is met, 0 otherwise.
# """
# def rule_out(x):
# for f in rule_out_callables:
# if f(x)==0:
# return 1
# return 0
#
# """
# 1 if any GI callable labels 1, -1 otherwise
# """
# def any_GI(x):
# for f in GI_callables:
# if f(x)==1:
# return 1
# return -1
#
# """
# 1 if any lab callable labels 1, -1 otherwise
# """
# def any_lab(x):
# for f in lab_callables:
# if f(x)==1:
# return 1
# return -1
#
# """
# Form LFs
# """
# for f in GI_callables:
# def new_lf(x, f=f):
# if f(x) == 1 and rule_out(x) == 0:
# return 1
# else:
# return -1
# new_lf.__name__ = f.__name__
# lfs.append(new_lf)
#
# def GI_any_lab(x, new_lf=new_lf):
# if new_lf(x) == 0:
# return 0
# elif new_lf(x) == 1 and any_lab(x) == 1:
# return 1
# else:
# return -1
# GI_any_lab.__name__ = f.__name__+'_lab'
# lfs.append(GI_any_lab)
#
#
# for f in lab_callables:
# def new_lf(x, f=f):
# if f(x) == 1 and rule_out(x) == 0:
# return 1
# else:
# return -1
# new_lf.__name__ = f.__name__
# lfs.append(new_lf)
#
# def lab_any_GI(x, new_lf=new_lf):
# if new_lf(x) == 0:
# return 0
# elif new_lf(x) == 1 and any_GI(x) == 1:
# return 1
# else:
# return -1
# lab_any_GI.__name__ = f.__name__+'_GI'
# lfs.append(lab_any_GI)
wrapped_lfs = []
for lf in lfs:
wrapped_lfs.append(LabelingFunction(name=lf.__name__, f=lf))
return wrapped_lfs
def make_lexicon_lf(thresh, pref="", previous_threshold=-sys.maxsize):
return LabelingFunction(
name="%s_less_%s" % (pref, thresh),
f=less_than_treshold,
resources=dict(thresh=thresh, previous_threshold=previous_threshold),
)
def make_keyword_lf(name, keywords_pos, keywords_neg):
return LabelingFunction(
name=name,
f=keyword_lookup,
resources=dict(keywords_pos=keywords_pos, keywords_neg=keywords_neg),
)
def make_expression_lf(name, pre_pos, expression):
return LabelingFunction(
name=name,
f=expression_lookup,
resources=dict(pre_pos=pre_pos, expression=expression),
)
def make_worker_lf(worker_id):
worker_dict = worker_dicts[worker_id]
name = f"worker_{worker_id}"
return LabelingFunction(name,
f=worker_lf,
resources={"worker_dict": worker_dict})
