def LF_part_ce_keywords_in_rows_cols_prefix(c):
ngrams = set(list(get_row_ngrams(c[1], n_max=3)))
ngrams = ngrams.union(set(list(get_col_ngrams(c[1], n_max=3))))
ngrams_part = _filter_non_parts(ngrams)
return (TRUE if overlap(_CE_KEYWORDS.union(_CE_ABBREVS), ngrams) and any(
[c.part.context.get_span().lower().startswith(_)
for _ in ngrams_part]) else ABSTAIN)
def LF_ce_keywords_not_part_in_row_col_prefix(c):
ngrams_part = set(list(get_col_ngrams(c[1], n_max=3, lower=False)))
ngrams_part = _filter_non_parts(
ngrams_part.union(set(list(get_row_ngrams(c[1], n_max=3,
lower=False)))))
return (TRUE if not same_table(c) and overlap(
_CE_KEYWORDS.union(_CE_ABBREVS), get_row_ngrams(c[1], n_max=3))
and len(ngrams_part) == 0
and not overlap(_NON_CEV_KEYWORDS, get_row_ngrams(c.part, n_max=3))
and not overlap(_NON_CEV_KEYWORDS, get_row_ngrams(c[1], n_max=3))
and not LF_current_in_row(c) else ABSTAIN)
def get_row_and_column_ngrams(mention):
row_ngrams = list(get_row_ngrams(mention))
col_ngrams = list(get_col_ngrams(mention))
if not mention.sentence.is_tabular():
assert len(row_ngrams) == 1 and row_ngrams[0] is None
assert len(col_ngrams) == 1 and col_ngrams[0] is None
else:
assert not any(x is None for x in row_ngrams)
assert not any(x is None for x in col_ngrams)
if "birth_place" in row_ngrams:
return True
else:
return False
def tablelib_unary_features(span):
"""
Table-/structure-related features for a single span
"""
if not span.sentence.is_tabular():
return
sentence = span.sentence
for attrib in settings["featurization"]["table"]["unary_features"]["attrib"]:
for ngram in get_cell_ngrams(
span,
n_max=settings["featurization"]["table"]["unary_features"][
"get_cell_ngrams"
]["max"],
attrib=attrib,
):
yield f"CELL_{attrib.upper()}_[{ngram}]", DEF_VALUE
for row_num in range(sentence.row_start, sentence.row_end + 1):
yield f"ROW_NUM_[{row_num}]", DEF_VALUE
for col_num in range(sentence.col_start, sentence.col_end + 1):
yield f"COL_NUM_[{col_num}]", DEF_VALUE
# NOTE: These two features could be accounted for by HTML_ATTR in
# structural features
yield f"ROW_SPAN_[{num_rows(sentence)}]", DEF_VALUE
yield f"COL_SPAN_[{num_cols(sentence)}]", DEF_VALUE
for axis in ["row", "col"]:
for ngram in get_head_ngrams(
span,
axis,
n_max=settings["featurization"]["table"]["unary_features"][
"get_head_ngrams"
]["max"],
attrib=attrib,
):
yield f"{axis.upper()}_HEAD_{attrib.upper()}_[{ngram}]", DEF_VALUE
for ngram in get_row_ngrams(
span,
n_max=settings["featurization"]["table"]["unary_features"][
"get_row_ngrams"
]["max"],
attrib=attrib,
):
yield f"ROW_{attrib.upper()}_[{ngram}]", DEF_VALUE
for ngram in get_col_ngrams(
span,
n_max=settings["featurization"]["table"]["unary_features"][
"get_col_ngrams"
]["max"],
attrib=attrib,
):
yield f"COL_{attrib.upper()}_[{ngram}]", DEF_VALUE
def neg_gain_keywords_in_column(c):
col_ngrams = set(get_col_ngrams(c.gain, n_max=1, lower=True))
if overlap(
[
"max",
"min",
"test",
"condition",
"conditions",
"vgn",
"f",
"-3",
"db",
"dbc",
],
col_ngrams,
):
return FALSE
else:
return ABSTAIN
def _condition(attr):
if overlap(["condition", "conditions"], get_col_ngrams(attr, n_max=1)):
return False
return True
def LF_part_mismatch_col(c):
ngrams_part = _filter_non_parts(set(list(get_col_ngrams(c[1], n_max=1))))
return (ABSTAIN if len(ngrams_part) == 0 or any([
c.part.context.get_span().lower().startswith(_.lower())
for _ in ngrams_part
]) else FALSE)
def neg_current_keywords_in_column(c):
return (FALSE if overlap(
["over", "temperature", "vgn", "f", "-3", "db", "dbc", "min", "max"],
get_col_ngrams(c.supply_current, lower=True),
) else ABSTAIN)
def pos_current_typ(c):
return (TRUE if overlap(["typ", "typ."],
get_col_ngrams(c.supply_current, lower=True)) else
ABSTAIN)
