def test_recursive_entity():
data = anafora.AnaforaData()
entity = anafora.AnaforaEntity()
entity.id = "@1@"
data.annotations.append(entity)
entity.properties["self"] = entity
assert entity.is_self_referential()
assert data.annotations.find_self_referential().id == entity.id
data = anafora.AnaforaData()
a = anafora.AnaforaEntity()
a.id = "A"
data.annotations.append(a)
b = anafora.AnaforaEntity()
b.id = "B"
data.annotations.append(b)
c = anafora.AnaforaEntity()
c.id = "C"
data.annotations.append(c)
d = anafora.AnaforaEntity()
d.id = "D"
data.annotations.append(d)
b.properties["x"] = a
c.properties["y"] = a
d.properties["1"] = b
d.properties["2"] = c
assert not d.is_self_referential()
def test_empty():
data = anafora.AnaforaData(anafora.ElementTree.fromstring('<data/>'))
assert list(data.annotations) == []
data = anafora.AnaforaData(
anafora.ElementTree.fromstring(
'<data><annotations></annotations></data>'))
assert list(data.annotations) == []
def test_missing_ignored_properties():
reference = anafora.AnaforaData(
anafora.ElementTree.fromstring("""
<data>
<annotations>
<entity>
<id>4</id>
<type>Z</type>
<properties>
<A>1</A>
<B>2</B>
<C></C>
</properties>
</entity>
</annotations>
</data>
"""))
predicted = anafora.AnaforaData(
anafora.ElementTree.fromstring("""
<data>
<annotations>
<entity>
<id>4</id>
<type>Z</type>
<properties>
<B>2</B>
<A>1</A>
</properties>
</entity>
</annotations>
</data>
"""))
named_scores = anafora.evaluate.score_data(reference, predicted)
scores = named_scores["Z"]
assert scores.correct == 0
assert scores.reference == 1
assert scores.predicted == 1
# make sure no exceptions are thrown
anafora.evaluate._print_document_scores([("temp", named_scores)])
named_scores = anafora.evaluate.score_data(reference,
predicted,
exclude=[("Z", "C")])
scores = named_scores["Z"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
def write(self, predictions):
"""Write predictions in anafora XML format"""
index = 0
if os.path.isdir(self.out_dir):
shutil.rmtree(self.out_dir)
os.mkdir(self.out_dir)
for sub_dir, text_name, file_names in \
anafora.walk(self.xml_dir, self.xml_regex):
xml_path = os.path.join(self.xml_dir, sub_dir, file_names[0])
ref_data = anafora.AnaforaData.from_file(xml_path)
data = anafora.AnaforaData()
for event in ref_data.annotations.select_type('EVENT'):
entity = anafora.AnaforaEntity()
entity.id = event.id
start, end = event.spans[0]
entity.spans = event.spans
entity.type = event.type
entity.properties['DocTimeRel'] = int2label[predictions[index]]
data.annotations.append(entity)
index = index + 1
os.mkdir(os.path.join(self.out_dir, sub_dir))
out_path = os.path.join(self.out_dir, sub_dir, file_names[0])
data.indent()
data.to_file(out_path)
def test_spans():
data = anafora.AnaforaData(
anafora.ElementTree.fromstring('''
<data>
<annotations>
<relation>
<id>1</id>
<type>R1</type>
<properties>
<relation>2</relation>
</properties>
</relation>
<relation>
<id>2</id>
<type>R2</type>
<properties>
<entity>3</entity>
</properties>
</relation>
<entity>
<id>3</id>
<type>E1</type>
<span>5,7</span>
</entity>
</annotations>
</data>'''))
assert data.annotations.select_id("1").spans == ((((5, 7), ), ), )
assert data.annotations.select_id("2").spans == (((5, 7), ), )
assert data.annotations.select_id("3").spans == ((5, 7), )
def test_add_entity():
data = anafora.AnaforaData()
assert str(data) == '<data />'
entity = anafora.AnaforaEntity()
with pytest.raises(ValueError) as exception_info:
data.annotations.append(entity)
assert "id" in str(exception_info.value)
assert str(data) == '<data />'
entity.id = "1"
data.annotations.append(entity)
assert str(
data
) == '<data><annotations><entity><id>1</id></entity></annotations></data>'
entity.type = "X"
entity.parents_type = "Y"
entity.properties["name1"] = "value1"
assert str(data) == ('<data><annotations><entity>' + '<id>1</id>' +
'<type>X</type>' + '<parentsType>Y</parentsType>' +
'<properties><name1>value1</name1></properties>' +
'</entity></annotations></data>')
del entity.properties["name1"]
assert str(data) == ('<data><annotations><entity>' + '<id>1</id>' +
'<type>X</type>' + '<parentsType>Y</parentsType>' +
'</entity></annotations></data>')
with pytest.raises(ValueError):
del entity.properties["name2"]
def prediction_to_anafora(model, labels, features, doc_name="dummy"):
data = anafora.AnaforaData()
for sent_labels, sent_features in zip(labels, features):
# Remove padding and <s> </s>
special_mask = model.tokenizer.get_special_tokens_mask(
sent_features.input_ids, already_has_special_tokens=True)
non_specials = np.count_nonzero(np.array(special_mask) == 0)
sent_labels = sent_labels[1:non_specials + 1]
sent_offsets = sent_features.offset_mapping[1:non_specials + 1]
previous_label = 0
previous_offset = [None, None]
for token_label, token_offset in zip(sent_labels, sent_offsets):
entity_label = model.config.id2label[
previous_label] if previous_label > 0 else None
new_word = token_offset[0] != previous_offset[
1] if model.config.pad_labels else True
label_diff = token_label - previous_label
if is_b_label(token_label, label_diff, model.config.bio_mode):
add_entity(data, doc_name, entity_label, previous_offset)
previous_label = token_label
previous_offset = token_offset
elif is_i_label(token_label, label_diff, new_word,
model.config.bio_mode):
previous_offset[1] = token_offset[1]
elif previous_label > 0:
add_entity(data, doc_name, entity_label, previous_offset)
previous_label = 0
previous_offset = [None, None]
if previous_label > 0:
entity_label = model.config.id2label[previous_label]
add_entity(data, doc_name, entity_label, previous_offset)
return data
def parse_json(input_dict):
data = anafora.AnaforaData()
list_of_entity_index = get_entity_index(input_dict)
for i in range(len(list_of_entity_index) - 1):
entity = build_an_entity(input_dict[list_of_entity_index[i]])
# when an entity has more than one group of span
if i < len(list_of_entity_index)-2 \
and input_dict[list_of_entity_index[i]+1]["type"] == "labels"\
and input_dict[list_of_entity_index[i]+1]["id"].split("@")[0].endswith("_continued"):
multiple_spans(entity, input_dict[list_of_entity_index[i] + 1])
entity.properties = set_up_properties(input_dict,
list_of_entity_index[i] + 2,
list_of_entity_index[i + 1],
entity)
else:
entity.properties = set_up_properties(input_dict,
list_of_entity_index[i] + 1,
list_of_entity_index[i + 1],
entity)
data.annotations.append(entity)
data.indent()
return data
def test_duplicate_id():
with pytest.raises(ValueError):
anafora.AnaforaData(
anafora.ElementTree.fromstring('''
<data>
<annotations>
<entity><id>1</id></entity>
<entity><id>1</id></entity>
</annotations>
</data>'''))
data = anafora.AnaforaData()
entity1 = anafora.AnaforaEntity()
entity1.id = "1"
entity2 = anafora.AnaforaEntity()
entity2.id = "1"
data.annotations.append(entity1)
with pytest.raises(ValueError):
data.annotations.append(entity2)
def test_add_entity():
"""Test testing testing"""
data = anafora.AnaforaData()
entity = anafora.AnaforaEntity()
entity.id = '1@e@ID025_path_074@gold'
data.annotations.append(entity)
entity.type = 'EVENT'
entity.parents_type = 'TemporalEntities'
entity.properties['DocTimeRel'] = 'AFTER'
data.indent()
data.to_file('temp.xml')
def span2xmlfiles(data_spans, file_name_simple):
import anafora
data = anafora.AnaforaData()
id = 0
for data_span in data_spans:
e = anafora.AnaforaEntity()
e.spans = ((int(data_span[0]), int(data_span[1]) + 1), )
e.type = data_span[2]
e.id = str(id) + "@e@" + file_name_simple
data.annotations.append(e)
id += 1
data.indent()
return data
def write_anafora(output_dir, dataset, predictions, tokenizer, config):
def add_entity(data, doc_name, label, offset):
entity_label = config.id2label[label] if label > 0 else None
if entity_label is not None:
anafora.AnaforaEntity()
entity = anafora.AnaforaEntity()
num_entities = len(data.xml.findall("annotations/entity"))
entity.id = "%s@%s" % (num_entities, doc_name)
entity.spans = ((offset[0], offset[1]),)
entity.type = entity_label.replace("B-", "")
data.annotations.append(entity)
for doc_index in dataset.doc_indices:
doc_subdir, doc_start, doc_end = doc_index
doc_name = os.path.basename(doc_subdir)
doc_features = dataset.features[doc_start:doc_end]
doc_predictions = predictions[doc_start:doc_end]
doc_predictions = np.argmax(doc_predictions, axis=2)
data = anafora.AnaforaData()
for sent_labels, sent_features in zip(doc_predictions, doc_features):
# Remove padding and <s> </s>
special_mask = tokenizer.get_special_tokens_mask(sent_features.input_ids,
already_has_special_tokens=True)
non_specials = np.count_nonzero(np.array(special_mask) == 0)
sent_labels = sent_labels[1: non_specials + 1]
sent_offsets = sent_features.offset_mapping[1: non_specials + 1]
previous_label = 0
previous_offset = [None, None] # (start, end)
for token_label, token_offset in zip(sent_labels, sent_offsets):
label_diff = token_label - previous_label
if token_label % 2 != 0: # If odd number, it is B label
add_entity(data, doc_name, previous_label, previous_offset)
previous_label = token_label
previous_offset = token_offset
elif label_diff == 1: # If even number and diff with previous is 1, it is I label
previous_offset[1] = token_offset[1]
elif previous_label > 0: # If current is O label and previous not O we must write it.
add_entity(data, doc_name, previous_label, previous_offset)
previous_label = 0
previous_offset = [None, None]
if previous_label > 0: # If remaining previous not O we must write it.
entity_label = config.id2label[previous_label]
add_entity(data, doc_name, entity_label, previous_offset)
doc_path = os.path.join(output_dir, doc_subdir)
os.makedirs(doc_path, exist_ok=True)
doc_path = os.path.join(doc_path,
"%s.TimeNorm.system.completed.xml" % doc_name)
print(doc_path)
data.to_file(doc_path)
def test_many_groups():
regex_predictions = {}
for i in range(1, 1000):
regex_predictions['a' * i] = ('A' * i, {})
annotator = anafora.regex.RegexAnnotator(regex_predictions)
text = "aaaaaaaaaa"
data = anafora.AnaforaData()
annotator.annotate(text, data)
assert len(list(data.annotations)) == 1
[annotation] = data.annotations
assert annotation.type == "AAAAAAAAAA"
assert annotation.spans == ((0, 10), )
assert dict(annotation.properties.items()) == {}
def write_xml(self, predicted_relations):
"""Write predictions in anafora XML format"""
# make a directory to write anafora xml
if os.path.isdir(self.out_dir):
shutil.rmtree(self.out_dir)
os.mkdir(self.out_dir)
# key: note, value: list of rel arg tuples
note2rels = defaultdict(list)
# map notes to relations in these notes
# for container_id, contained_id in predicted_relations:
for contained_id, container_id in predicted_relations:
note_name = container_id.split('@')[2]
note2rels[note_name].append((container_id, contained_id))
# iterate over reference anafora xml files
for sub_dir, text_name, file_names in anafora.walk(
self.xml_dir, self.xml_regex):
path = os.path.join(self.xml_dir, sub_dir, file_names[0])
ref_data = anafora.AnaforaData.from_file(path)
# make a new XML file
generated_data = anafora.AnaforaData()
# copy gold events and times
copy_annotations(ref_data, generated_data, 'EVENT')
copy_annotations(ref_data, generated_data, 'TIMEX3')
copy_annotations(ref_data, generated_data, 'SECTIONTIME')
copy_annotations(ref_data, generated_data, 'DOCTIME')
# add generated relations
note_name = file_names[0].split('.')[0]
for container_id, contained_id in note2rels[note_name]:
relation = anafora.AnaforaRelation()
relation.id = str(random.random())[2:]
relation.type = 'TLINK'
relation.parents_type = 'TemporalRelations'
relation.properties['Source'] = container_id
relation.properties['Type'] = 'CONTAINS'
relation.properties['Target'] = contained_id
generated_data.annotations.append(relation)
# write xml to file
generated_data.indent()
os.mkdir(os.path.join(self.out_dir, sub_dir))
out_path = os.path.join(self.out_dir, sub_dir, file_names[0])
generated_data.to_file(out_path)
def _annotate(model_file,
text_dir,
output_dir,
data_dir=None,
xml_name_regex="[.]xml$",
text_encoding="utf-8",
extension=".system.completed.xml"):
if text_dir is not None:
iterator = anafora.walk_flat_to_anafora(text_dir)
elif data_dir is not None:
iterator = anafora.walk_anafora_to_anafora(data_dir, xml_name_regex)
else:
iterator = anafora.walk_anafora_to_anafora(output_dir, xml_name_regex)
# load a model from the file
model = RegexAnnotator.from_file(model_file)
# annotate each text
for input_sub_dir, output_sub_dir, text_name, xml_names in iterator:
if data_dir is None:
data_iter = [(anafora.AnaforaData(), text_name + extension)]
else:
data_iter = [(anafora.AnaforaData.from_file(
os.path.join(data_dir, input_sub_dir, xml_name)),
regex.sub(r'[.][^.]*[.][^.]*[.][^.]*[.]xml',
extension, xml_name))
for xml_name in xml_names]
for data, output_name in data_iter:
# read in the text
if text_dir is not None:
text_path = os.path.join(text_dir, text_name)
elif data_dir is not None:
text_path = os.path.join(data_dir, input_sub_dir, text_name)
else:
text_path = os.path.join(output_dir, input_sub_dir, text_name)
with codecs.open(text_path, 'r', text_encoding) as text_file:
text = text_file.read()
# annotate the text
model.annotate(text, data)
# save the annotated data to the output directory
data_output_dir = os.path.join(output_dir, output_sub_dir)
if not os.path.exists(data_output_dir):
os.makedirs(data_output_dir)
data_output_path = os.path.join(data_output_dir, output_name)
data.indent()
data.to_file(data_output_path)
def write_xml(self, prediction_lookup):
"""Write predictions in anafora XML format"""
# make a directory to write anafora xml
if os.path.isdir(self.xml_out_dir):
shutil.rmtree(self.xml_out_dir)
os.mkdir(self.xml_out_dir)
# t5 occasionally fails to predict
missing_predictions = []
# iterate over reference xml files
# look up the DTR prediction for each event
# and write it in anafora format to specificed dir
for sub_dir, text_name, file_names in \
anafora.walk(self.xml_ref_dir, xml_regex):
path = os.path.join(self.xml_ref_dir, sub_dir, file_names[0])
ref_data = anafora.AnaforaData.from_file(path)
data = anafora.AnaforaData()
for event in ref_data.annotations.select_type('EVENT'):
# make a new entity and copy some ref info
entity = anafora.AnaforaEntity()
entity.id = event.id
start, end = event.spans[0]
entity.spans = event.spans
entity.type = event.type
# lookup the prediction
key = '|'.join((sub_dir, str(start), str(end)))
if key not in prediction_lookup:
# use majority class for now
entity.properties['DocTimeRel'] = 'OVERLAP'
missing_predictions.append(key)
else:
entity.properties['DocTimeRel'] = prediction_lookup[key]
data.annotations.append(entity)
data.indent()
os.mkdir(os.path.join(self.xml_out_dir, sub_dir))
out_path = os.path.join(self.xml_out_dir, sub_dir, file_names[0])
data.to_file(out_path)
print('number of missing predictions:', len(missing_predictions))
def test_filter_by_precision():
annotator = anafora.regex.RegexAnnotator({
r'the': ("THE", {}),
r'\bthe\b': ("THE", {}),
r'yer\b': ("ER", {}),
r'er\b': ("ER", {})
})
text = "the theater near the record player"
data = anafora.AnaforaData(
anafora.ElementTree.fromstring("""
<data>
<annotations>
<entity>
<id>1</id>
<type>THE</type>
<span>0,3</span><!-- "the" -->
</entity>
<entity>
<id>2</id>
<type>THE</type>
<span>17,20</span><!-- "the" -->
</entity>
<entity>
<id>3</id>
<type>ER</type>
<span>9,11</span><!-- "er" -->
</entity>
<entity>
<id>4</id>
<type>ER</type>
<span>32,34</span><!-- "." -->
</entity>
</annotations>
</data>
"""))
annotator.prune_by_precision(0.6, [(text, data)])
assert annotator == anafora.regex.RegexAnnotator({
r'the': ("THE", {}),
r'\bthe\b': ("THE", {}),
r'er\b': ("ER", {})
})
annotator.prune_by_precision(1.0, [(text, data)])
assert annotator == anafora.regex.RegexAnnotator({
r'\bthe\b': ("THE", {}),
r'er\b': ("ER", {})
})
def test_sort():
data = anafora.AnaforaData(
anafora.ElementTree.fromstring('''
<data>
<annotations>
<entity>
<id>1</id>
<type>E</type>
<span>5,7</span>
</entity>
<entity>
<id>2</id>
<type>E</type>
<span>3,4</span>
</entity>
</annotations>
</data>'''))
assert [a.id for a in sorted(data.annotations)] == ['2', '1']
def write(self, predictions):
"""Write predictions in anafora XML format"""
# predictions are in the same order in which they were read
prediction_lookup = dict(zip(self.offsets, predictions))
# make a directory to write anafora xml
if os.path.isdir(self.xml_out_dir):
shutil.rmtree(self.xml_out_dir)
os.mkdir(self.xml_out_dir)
# iterate over reference xml files
# look up the DTR prediction for each event
# and write it in anafora format to specificed dir
for sub_dir, text_name, file_names in \
anafora.walk(self.xml_ref_dir, xml_regex):
path = os.path.join(self.xml_ref_dir, sub_dir, file_names[0])
ref_data = anafora.AnaforaData.from_file(path)
data = anafora.AnaforaData()
for event in ref_data.annotations.select_type('EVENT'):
# make a new entity and copy some ref info
entity = anafora.AnaforaEntity()
entity.id = event.id
start, end = event.spans[0]
entity.spans = event.spans
entity.type = event.type
# lookup the prediction
if (sub_dir, start, end) not in prediction_lookup:
print('missing key:', (sub_dir, start, end))
continue
label = prediction_lookup[(sub_dir, start, end)]
entity.properties['DocTimeRel'] = int2label[label]
data.annotations.append(entity)
data.indent()
os.mkdir(os.path.join(self.xml_out_dir, sub_dir))
out_path = os.path.join(self.xml_out_dir, sub_dir, file_names[0])
data.to_file(out_path)
def test_add_reference():
data = anafora.AnaforaData()
entity1 = anafora.AnaforaEntity()
entity1.id = "@1@"
entity2 = anafora.AnaforaEntity()
entity2.id = "@2@"
with pytest.raises(ValueError) as exception_info:
entity2.properties["link"] = entity1
assert "<annotations" in str(exception_info.value)
data.annotations.append(entity1)
with pytest.raises(ValueError):
entity2.properties["link"] = entity1
assert "<annotations" in str(exception_info.value)
data.annotations.append(entity2)
entity2.properties["link"] = entity1
assert str(data) == (
'<data><annotations>' + '<entity><id>@1@</id></entity>' +
'<entity><id>@2@</id><properties><link>@1@</link></properties></entity>'
+ '</annotations></data>')
def test_regex_annotator():
annotator = anafora.regex.RegexAnnotator({
'aa+': ('A', {}),
'a': ('A', {
'X': '2'
}),
'bb': ('B', {
'Y': '1'
})
})
text = "bb aaa"
data = anafora.AnaforaData()
annotator.annotate(text, data)
assert len(list(data.annotations)) == 2
[b_annotation, a_annotation] = data.annotations
assert b_annotation.type == "B"
assert b_annotation.spans == ((0, 2), )
assert dict(b_annotation.properties.items()) == {'Y': '1'}
assert a_annotation.type == "A"
assert a_annotation.spans == ((3, 6), )
assert dict(a_annotation.properties.items()) == {}
def test_preannotated():
annotator = anafora.regex.RegexAnnotator(
{
'aa+': ('A', {
'X': '2'
}),
'a': ('A', {}),
'bb': ('B', {
'Y': '1'
})
}, {'C': {
'Z': '3'
}})
text = "bb aaa"
data = anafora.AnaforaData()
bb = anafora.AnaforaEntity()
bb.id = "1@preannotated"
bb.type = "B"
bb.spans = ((0, 2), )
data.annotations.append(bb)
aaa = anafora.AnaforaEntity()
aaa.id = "2@preannotated"
aaa.type = "C"
aaa.spans = ((3, 6), )
data.annotations.append(aaa)
annotator.annotate(text, data)
assert len(list(data.annotations)) == 3
[b_annotation, c_annotation, a_annotation] = data.annotations
assert b_annotation.type == "B"
assert b_annotation.spans == ((0, 2), )
assert dict(b_annotation.properties.items()) == {'Y': '1'}
assert c_annotation.type == "C"
assert c_annotation.spans == ((3, 6), )
assert dict(c_annotation.properties.items()) == {'Z': '3'}
assert a_annotation.type == "A"
assert a_annotation.spans == ((3, 6), )
assert dict(a_annotation.properties.items()) == {'X': '2'}
def test_remove():
data = anafora.AnaforaData()
assert str(data) == '<data />'
entity1 = anafora.AnaforaEntity()
entity1.id = "@1@"
data.annotations.append(entity1)
entity2 = anafora.AnaforaEntity()
entity2.id = "@2@"
entity2.properties["name"] = "value"
data.annotations.append(entity2)
assert list(data.annotations) == [entity1, entity2]
assert str(data) == (
'<data><annotations>' + '<entity><id>@1@</id></entity>' +
'<entity><id>@2@</id><properties><name>value</name></properties></entity>'
+ '</annotations></data>')
data.annotations.remove(entity1)
assert list(data.annotations) == [entity2]
assert str(data) == (
'<data><annotations>' +
'<entity><id>@2@</id><properties><name>value</name></properties></entity>'
+ '</annotations></data>')
data.annotations.remove(entity2)
assert list(data.annotations) == []
assert str(data) == '<data><annotations /></data>'
def span2xmlfiles(exp, target):
import anafora
raw_dir_simple = read1.read_from_json('raw_dir_simple')
for data_id in range(0, 10):
data_spans = read1.read_json(exp + "\\span_label_all" +
target)[data_id]
data = anafora.AnaforaData()
id = 0
for data_span in data_spans:
e = anafora.AnaforaEntity()
e.spans = ((int(data_span[0]), int(data_span[1]) + 1), )
e.type = data_span[2]
e.id = str(id) + "@e@" + raw_dir_simple[data_id]
data.annotations.append(e)
id += 1
print data
data.indent()
outputfile = exp + "\\" + raw_dir_simple[data_id] + "\\"
if not os.path.exists(outputfile):
os.makedirs(outputfile)
data.to_file(outputfile + raw_dir_simple[data_id] +
".TimeNorm.gold.completed.xml")
def test_score_data():
reference = anafora.AnaforaData(
anafora.ElementTree.fromstring("""
<data>
<annotations>
<entity>
<id>1</id>
<span>0,5</span>
<type>X</type>
</entity>
<entity>
<id>2</id>
<span>5,10</span>
<type>Y</type>
</entity>
<entity>
<id>3</id>
<span>15,20</span>
<type>Y</type>
</entity>
<relation>
<id>4</id>
<type>Z</type>
<properties>
<Source>1</Source>
<Target>2</Target>
<Prop1>T</Prop1>
<Prop2>A</Prop2>
</properties>
</relation>
<relation>
<id>5</id>
<type>Z</type>
<properties>
<Source>2</Source>
<Target>3</Target>
<Prop1>T</Prop1>
<Prop2>B</Prop2>
</properties>
</relation>
</annotations>
</data>
"""))
predicted = anafora.AnaforaData(
anafora.ElementTree.fromstring("""
<data>
<annotations>
<entity>
<id>6</id><!-- different -->
<span>0,5</span>
<type>X</type>
</entity>
<entity>
<id>7</id><!-- different -->
<span>5,10</span>
<type>X</type><!-- different -->
</entity>
<entity>
<id>8</id><!-- different -->
<span>15,20</span>
<type>Y</type>
</entity>
<relation>
<id>9</id><!-- different -->
<type>Z</type>
<properties>
<Source>6</Source>
<Target>7</Target>
<Prop1>T</Prop1>
<Prop2>A</Prop2>
</properties>
</relation>
<relation>
<id>10</id><!-- different -->
<type>Z</type>
<properties>
<Source>7</Source>
<Target>8</Target>
<Prop1>F</Prop1><!-- different -->
<Prop2>B</Prop2>
</properties>
</relation>
</annotations>
</data>
"""))
named_scores = anafora.evaluate.score_data(reference, predicted)
assert set(named_scores.keys()) == {
"*",
("*", "<span>"),
"X",
("X", "<span>"),
"Y",
("Y", "<span>"),
"Z",
("Z", "<span>"),
("Z", "Source"),
("Z", "Target"),
("Z", "Prop1"),
("Z", "Prop2"),
("Z", "Prop1", "T"),
("Z", "Prop1", "F"),
("Z", "Prop2", "A"),
("Z", "Prop2", "B"),
}
scores = named_scores["X"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 2
scores = named_scores["X", "<span>"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 2
scores = named_scores["Y"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 1
scores = named_scores["Y", "<span>"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 1
scores = named_scores["Z"]
assert scores.correct == 0
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "<span>"]
assert scores.correct == 2
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop1"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop1", "T"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 1
scores = named_scores["Z", "Prop1", "F"]
assert scores.correct == 0
assert scores.reference == 0
assert scores.predicted == 1
scores = named_scores["Z", "Prop2"]
assert scores.correct == 2
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop2", "A"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["Z", "Prop2", "B"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["*"]
assert scores.correct == 1 + 1 + 0
assert scores.reference == 1 + 2 + 2
assert scores.predicted == 2 + 1 + 2
scores = named_scores["*", "<span>"]
assert scores.correct == 1 + 1 + 2
assert scores.reference == 1 + 2 + 2
assert scores.predicted == 2 + 1 + 2
named_scores = anafora.evaluate.score_data(reference,
predicted,
exclude=["X", "Y"])
assert set(named_scores.keys()) == {
"*",
("*", "<span>"),
"Z",
("Z", "<span>"),
("Z", "Source"),
("Z", "Target"),
("Z", "Prop1"),
("Z", "Prop2"),
("Z", "Prop1", "T"),
("Z", "Prop1", "F"),
("Z", "Prop2", "A"),
("Z", "Prop2", "B"),
}
scores = named_scores["Z"]
assert scores.correct == 0
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "<span>"]
assert scores.correct == 2
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop1"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop1", "T"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 1
scores = named_scores["Z", "Prop1", "F"]
assert scores.correct == 0
assert scores.reference == 0
assert scores.predicted == 1
scores = named_scores["Z", "Prop2"]
assert scores.correct == 2
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop2", "A"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["Z", "Prop2", "B"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["*"]
assert scores.correct == 0
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["*", "<span>"]
assert scores.correct == 2
assert scores.reference == 2
assert scores.predicted == 2
named_scores = anafora.evaluate.score_data(reference,
predicted,
include=[("Z", "Prop1", "T")])
assert set(named_scores.keys()) == {("Z", "Prop1", "T")}
scores = named_scores["Z", "Prop1", "T"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 1
named_scores = anafora.evaluate.score_data(reference,
predicted,
include=[("Z", "Prop1", "F")])
assert set(named_scores.keys()) == {("Z", "Prop1", "F")}
scores = named_scores["Z", "Prop1", "F"]
assert scores.correct == 0
assert scores.reference == 0
assert scores.predicted == 1
named_scores = anafora.evaluate.score_data(reference,
predicted,
include=["Z"],
exclude=[("Z", "<span>")])
assert set(named_scores.keys()) == {
"*",
"Z",
("Z", "Source"),
("Z", "Target"),
("Z", "Prop1"),
("Z", "Prop2"),
("Z", "Prop1", "T"),
("Z", "Prop1", "F"),
("Z", "Prop2", "A"),
("Z", "Prop2", "B"),
}
scores = named_scores["Z"]
assert scores.correct == 0
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop1"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop1", "T"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 1
scores = named_scores["Z", "Prop1", "F"]
assert scores.correct == 0
assert scores.reference == 0
assert scores.predicted == 1
scores = named_scores["Z", "Prop2"]
assert scores.correct == 2
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop2", "A"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["Z", "Prop2", "B"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["*"]
assert scores.correct == 0
assert scores.reference == 2
assert scores.predicted == 2
def test_score_data_overlap():
# This test is identical to the one above except that the spans have been changed so that they're overlapping
# instead of being exactly equal
reference = anafora.AnaforaData(
anafora.ElementTree.fromstring("""
<data>
<annotations>
<entity>
<id>1</id>
<span>0,5</span>
<type>X</type>
</entity>
<entity>
<id>2</id>
<span>5,10</span>
<type>Y</type>
</entity>
<entity>
<id>3</id>
<span>15,20</span>
<type>Y</type>
</entity>
<relation>
<id>4</id>
<type>Z</type>
<properties>
<Source>1</Source>
<Target>2</Target>
<Prop1>T</Prop1>
<Prop2>A</Prop2>
</properties>
</relation>
<relation>
<id>5</id>
<type>Z</type>
<properties>
<Source>2</Source>
<Target>3</Target>
<Prop1>T</Prop1>
<Prop2>B</Prop2>
</properties>
</relation>
<relation>
<id>6</id>
<type>Ref</type>
<properties>
<Ref>1</Ref>
</properties>
</relation>
<relation>
<id>7</id>
<type>Ref</type>
<properties>
<Ref>6</Ref>
</properties>
</relation>
</annotations>
</data>
"""))
predicted = anafora.AnaforaData(
anafora.ElementTree.fromstring("""
<data>
<annotations>
<entity>
<id>6</id><!-- different -->
<span>0,4</span>
<type>X</type>
</entity>
<entity>
<id>7</id><!-- different -->
<span>6,10</span>
<type>X</type><!-- different -->
</entity>
<entity>
<id>8</id><!-- different -->
<span>19,20</span>
<type>Y</type>
</entity>
<relation>
<id>9</id><!-- different -->
<type>Z</type>
<properties>
<Source>6</Source>
<Target>7</Target>
<Prop1>T</Prop1>
<Prop2>A</Prop2>
</properties>
</relation>
<relation>
<id>10</id><!-- different -->
<type>Z</type>
<properties>
<Source>7</Source>
<Target>8</Target>
<Prop1>F</Prop1><!-- different -->
<Prop2>B</Prop2>
</properties>
</relation>
</annotations>
</data>
"""))
named_scores = anafora.evaluate.score_data(
reference,
predicted,
annotation_wrapper=anafora.evaluate._OverlappingWrapper)
assert set(named_scores.keys()) == {
"*",
("*", "<span>"),
"X",
("X", "<span>"),
"Y",
("Y", "<span>"),
"Z",
("Z", "<span>"),
("Z", "Source"),
("Z", "Target"),
("Z", "Prop1"),
("Z", "Prop2"),
("Z", "Prop1", "T"),
("Z", "Prop1", "F"),
("Z", "Prop2", "A"),
("Z", "Prop2", "B"),
"Ref",
("Ref", "<span>"),
("Ref", "Ref"),
}
scores = named_scores["X"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 2
scores = named_scores["X", "<span>"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 2
scores = named_scores["Y"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 1
scores = named_scores["Y", "<span>"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 1
scores = named_scores["Z"]
assert scores.correct == 0
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "<span>"]
assert scores.correct == 2
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop1"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop1", "T"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 1
scores = named_scores["Z", "Prop1", "F"]
assert scores.correct == 0
assert scores.reference == 0
assert scores.predicted == 1
scores = named_scores["Z", "Prop2"]
assert scores.correct == 2
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop2", "A"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["Z", "Prop2", "B"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["*"]
assert scores.correct == 1 + 1 + 0
assert scores.reference == 1 + 2 + 2 + 2
assert scores.predicted == 2 + 1 + 2
scores = named_scores["*", "<span>"]
assert scores.correct == 1 + 1 + 2
assert scores.reference == 1 + 2 + 2 + 1
assert scores.predicted == 2 + 1 + 2
named_scores = anafora.evaluate.score_data(
reference,
predicted,
exclude=["X", "Y"],
annotation_wrapper=anafora.evaluate._OverlappingWrapper)
assert set(named_scores.keys()) == {
"*",
("*", "<span>"),
"Z",
("Z", "<span>"),
("Z", "Source"),
("Z", "Target"),
("Z", "Prop1"),
("Z", "Prop2"),
("Z", "Prop1", "T"),
("Z", "Prop1", "F"),
("Z", "Prop2", "A"),
("Z", "Prop2", "B"),
"Ref",
("Ref", "<span>"),
("Ref", "Ref"),
}
scores = named_scores["Z"]
assert scores.correct == 0
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "<span>"]
assert scores.correct == 2
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop1"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop1", "T"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 1
scores = named_scores["Z", "Prop1", "F"]
assert scores.correct == 0
assert scores.reference == 0
assert scores.predicted == 1
scores = named_scores["Z", "Prop2"]
assert scores.correct == 2
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop2", "A"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["Z", "Prop2", "B"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["*"]
assert scores.correct == 0
assert scores.reference == 2 + 2
assert scores.predicted == 2
scores = named_scores["*", "<span>"]
assert scores.correct == 2
assert scores.reference == 2 + 1
assert scores.predicted == 2
named_scores = anafora.evaluate.score_data(
reference,
predicted,
include=[("Z", "Prop1", "T")],
annotation_wrapper=anafora.evaluate._OverlappingWrapper)
assert set(named_scores.keys()) == {("Z", "Prop1", "T")}
scores = named_scores["Z", "Prop1", "T"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 1
named_scores = anafora.evaluate.score_data(
reference,
predicted,
include=[("Z", "Prop1", "F")],
annotation_wrapper=anafora.evaluate._OverlappingWrapper)
assert set(named_scores.keys()) == {("Z", "Prop1", "F")}
scores = named_scores["Z", "Prop1", "F"]
assert scores.correct == 0
assert scores.reference == 0
assert scores.predicted == 1
named_scores = anafora.evaluate.score_data(
reference,
predicted,
include=["Z"],
exclude=[("Z", "<span>")],
annotation_wrapper=anafora.evaluate._OverlappingWrapper)
assert set(named_scores.keys()) == {
"*",
"Z",
("Z", "Source"),
("Z", "Target"),
("Z", "Prop1"),
("Z", "Prop2"),
("Z", "Prop1", "T"),
("Z", "Prop1", "F"),
("Z", "Prop2", "A"),
("Z", "Prop2", "B"),
}
scores = named_scores["Z"]
assert scores.correct == 0
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop1"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop1", "T"]
assert scores.correct == 1
assert scores.reference == 2
assert scores.predicted == 1
scores = named_scores["Z", "Prop1", "F"]
assert scores.correct == 0
assert scores.reference == 0
assert scores.predicted == 1
scores = named_scores["Z", "Prop2"]
assert scores.correct == 2
assert scores.reference == 2
assert scores.predicted == 2
scores = named_scores["Z", "Prop2", "A"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["Z", "Prop2", "B"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["*"]
assert scores.correct == 0
assert scores.reference == 2
assert scores.predicted == 2
def score_dirs(reference_dir,
predicted_dir,
xml_name_regex="[.]xml$",
text_dir=None,
include=None,
exclude=None,
scores_type=Scores,
annotation_wrapper=None):
"""
:param string reference_dir: directory containing reference ("gold standard") Anafora XML directories
:param string predicted_dir: directory containing predicted (system-generated) Anafora XML directories
:param xml_name_regex: regular expression matching the files to be compared
:param string text_dir: directory containing the raw texts corresponding to the Anafora XML
(if None, texts are assumed to be in the reference dir)
:param set include: types of annotations to include (others will be excluded); may be type names,
(type-name, property-name) tuples, (type-name, property-name, property-value) tuples
:param set exclude: types of annotations to exclude; may be type names, (type-name, property-name) tuples,
(type-name, property-name, property-value) tuples
:param type scores_type: type for calculating matches between predictions and reference
:param type annotation_wrapper: wrapper object to apply to AnaforaAnnotations
:return iter: an iterator of (file-name, name-to-scores) where name-to-scores is a mapping from
(annotation type[, property name[, property value]]) to a Scores object
"""
# walks through the reference Anafora XML directories, scoring each and adding those to the overall scores
for sub_dir, text_name, reference_xml_names in anafora.walk(
reference_dir, xml_name_regex):
# load the reference data from its Anafora XML
try:
[reference_xml_name] = reference_xml_names
except ValueError:
logging.warn("expected one reference file for %s, found %s",
text_name, reference_xml_names)
if not reference_xml_names:
continue
reference_xml_name = reference_xml_names[0]
reference_xml_path = os.path.join(reference_dir, sub_dir,
reference_xml_name)
reference_data = _load(reference_xml_path)
# check for self-references in the annotations, which cause equality and hashing to fail
self_reference = reference_data.annotations.find_self_referential()
if self_reference is not None:
msg = "skipping reference file %s with self-referential annotation %s"
logging.warn(msg, reference_xml_path, self_reference.id)
continue
# find and load the corresponding predicted data from its Anafora XML
predicted_xml_glob = os.path.join(predicted_dir, sub_dir,
text_name + "*.xml")
predicted_xml_paths = [
f for f in glob.glob(predicted_xml_glob)
if re.search(xml_name_regex, f) is not None
]
try:
[predicted_xml_path] = predicted_xml_paths
predicted_data = _load(predicted_xml_path)
except ValueError:
logging.warn("expected one predicted file at %s, found %s",
predicted_xml_glob, predicted_xml_paths)
if not predicted_xml_paths:
predicted_xml_path = None
predicted_data = anafora.AnaforaData()
else:
predicted_xml_path = predicted_xml_paths[0]
predicted_data = _load(predicted_xml_path)
# check for self-references in the annotations, which cause equality and hashing to fail
self_reference = predicted_data.annotations.find_self_referential()
if self_reference is not None:
msg = "skipping predicted file %s with self-referential annotation %s"
logging.warn(msg, predicted_xml_path, self_reference.id)
predicted_data = anafora.AnaforaData()
# determine the path for the raw text source file
if text_dir is None:
text_path = os.path.join(reference_dir, sub_dir, text_name)
else:
text_path = os.path.join(text_dir, text_name)
# if no raw text was found, then asking for the text of an annotation is an error
if not os.path.exists(text_path) or not os.path.isfile(text_path):
def _span_text(_):
raise RuntimeError(
"no text file found at {0}".format(text_path))
# otherwise, the text of an annotation can be extracted based on its spans
else:
with open(text_path) as text_file:
text = text_file.read()
def _flatten(items):
if isinstance(items, tuple) and isinstance(items[0], int):
yield items
else:
for item in items:
for flattened_items in _flatten(item):
yield flattened_items
def _span_text(spans):
return "...".join(text[start:end]
for start, end in _flatten(spans))
# score this data and update the overall scores
named_scores = score_data(reference_data,
predicted_data,
include,
exclude,
scores_type=scores_type,
annotation_wrapper=annotation_wrapper)
for name, scores in named_scores.items():
# if there were some predictions, and if we're using scores that keep track of errors, log the errors
if predicted_xml_paths:
for annotation, message in getattr(scores, "errors", []):
logging.debug('%s: %s: "%s" %s"', text_name, message,
_span_text(annotation.spans), annotation)
# generate the file name and the resulting scores
yield text_name, named_scores
def test_schema_validate():
schema = anafora.validate.Schema(anafora.ElementTree.fromstring("""
<schema>
<defaultattribute>
<required>True</required>
</defaultattribute>
<definition>
<entities>
<entity type="X">
<properties>
<property type="A" input="choice">x,y</property>
<property type="B" />
<property type="C" instanceOf="Y,Z" />
</properties>
</entity>
<entity type="Y" />
<entity type="Z" />
</entities>
<relations>
<relation type="R">
<properties>
<property type="D" instanceOf="X" required="False" />
<property type="E" instanceOf="Y,Z" required="False" />
</properties>
</relation>
</relations>
</definition>
</schema>
"""))
data = anafora.AnaforaData()
entity1 = anafora.AnaforaEntity()
entity1.id = "@1@"
entity1.type = "X"
entity1.properties["A"] = "x"
data.annotations.append(entity1)
assert schema.errors(data)
entity1.properties["B"] = "y"
assert schema.errors(data)
entity1.properties["C"] = "z"
assert schema.errors(data)
entity2 = anafora.AnaforaEntity()
entity2.id = "@2@"
entity2.type = "X"
data.annotations.append(entity2)
entity1.properties["C"] = entity2
assert schema.errors(data)
entity2.type = "Y"
assert not schema.errors(data)
entity1.properties["A"] = "y"
assert not schema.errors(data)
entity1.properties["A"] = "z"
assert schema.errors(data)
entity1.properties["A"] = "x"
assert not schema.errors(data)
relation = anafora.AnaforaRelation()
relation.id = "@3@"
relation.type = ""
data.annotations.append(relation)
assert schema.errors(data)
relation.type = "R"
assert not schema.errors(data)
relation.properties["D"] = entity1
assert not schema.errors(data)
relation.properties["E"] = entity1
assert schema.errors(data)
relation.properties["E"] = entity2
assert not schema.errors(data)
relation.properties["X"] = "Y"
assert schema.errors(data)
def to_anafora_data(timeml_path):
"""
:param xml.etree.ElementTree.Element timeml_path: path of the TimeML XML
:return anafora.AnaforaData: an Anafora version of the TimeML annotations
"""
entity_tags = {"TIMEX3", "EVENT", "SIGNAL"}
tag_id_attrs = {
"TIMEX3": "tid",
"EVENT": "eid",
"SIGNAL": "sid",
"MAKEINSTANCE": "eiid",
"TLINK": "lid",
"SLINK": "lid",
"ALINK": "lid",
}
ref_id_attrs = {
"eventID", "signalID", "beginPoint", "endPoint", "valueFromFunction",
"anchorTimeID", "eventInstanceID", "timeID", "signalID",
"relatedToEventInstance", "relatedToTime", "subordinatedEventInstance",
"tagID"
}
text = to_text(timeml_path)
data = anafora.AnaforaData()
root = anafora.ElementTree.parse(timeml_path).getroot()
prefix_to_char = {'t': 'e', 'e': 'e', 's': 'e', 'ei': 'r', 'l': 'r'}
timeml_id_to_anafora_id = {}
count = 1
file_base, _ = os.path.splitext(os.path.basename(timeml_path))
for elem in root.iter():
if elem.tag in tag_id_attrs:
timeml_id = elem.attrib[tag_id_attrs[elem.tag]]
[(prefix, number)] = re.findall(r'^(\D+)(\d+)$', timeml_id)
timeml_id_to_anafora_id[timeml_id] = '{0:d}@{1}@{2}@gold'.format(
count, prefix_to_char[prefix], file_base)
count += 1
def add_annotations_from(elem, offset=0):
start = offset
annotation = None
if elem.tag in tag_id_attrs:
annotation = anafora.AnaforaEntity(
) if elem.tag in entity_tags else anafora.AnaforaRelation()
id_attr = tag_id_attrs[elem.tag]
annotation.id = timeml_id_to_anafora_id[elem.attrib[id_attr]]
annotation.type = elem.tag
if isinstance(annotation, anafora.AnaforaEntity):
annotation.spans = ((start, start), )
for name, value in elem.attrib.items():
if name != id_attr:
if name in ref_id_attrs:
value = timeml_id_to_anafora_id[value]
annotation.properties[name] = value
data.annotations.append(annotation)
if elem.text is not None:
offset += len(elem.text)
for child in elem:
offset = add_annotations_from(child, offset)
if annotation is not None and isinstance(annotation,
anafora.AnaforaEntity):
annotation.spans = ((start, offset), )
if elem.text != text[start:offset]:
raise ValueError('{0}: "{1}" != "{2}"'.format(
timeml_path, elem.text, text[start:offset]))
if elem.tail is not None:
offset += len(elem.tail)
return offset
add_annotations_from(root)
return data
def test_delete_excluded():
reference = anafora.AnaforaData(
anafora.ElementTree.fromstring("""
<data>
<annotations>
<entity>
<id>1@e</id>
<type>Z</type>
<span>1, 3</span>
<properties>
<A>2@e</A>
</properties>
</entity>
<entity>
<id>2@e</id>
<type>Y</type>
<span>4, 6</span>
<properties>
<B>3@e</B>
</properties>
</entity>
<entity>
<id>3@e</id>
<type>X</type>
<span>7, 9</span>
</entity>
<entity>
<id>4@e</id>
<type>W</type>
<span>20, 30</span>
<properties>
<B>3@e</B>
</properties>
</entity>
</annotations>
</data>
"""))
predicted = anafora.AnaforaData(
anafora.ElementTree.fromstring("""
<data>
<annotations>
<entity>
<id>4@e</id>
<type>Z</type>
<span>1, 3</span>
<properties>
<A>5@e</A>
</properties>
</entity>
<entity>
<id>5@e</id>
<type>Y</type>
<span>4, 6</span>
<properties>
<B>6@e</B>
</properties>
</entity>
<entity>
<id>6@e</id>
<type>X</type>
<span>10, 15</span>
</entity>
<entity>
<id>7@e</id>
<type>W</type>
<span>20, 30</span>
<properties>
<B></B>
</properties>
</entity>
</annotations>
</data>
"""))
named_scores = anafora.evaluate.score_data(reference, predicted)
scores = named_scores["X"]
assert scores.correct == 0
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["Y"]
assert scores.correct == 0
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["Z"]
assert scores.correct == 0
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["W"]
assert scores.correct == 0
assert scores.reference == 1
assert scores.predicted == 1
named_scores = anafora.evaluate.score_data(reference,
predicted,
exclude={"X"})
scores = named_scores["Y"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["Z"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["W"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
named_scores = anafora.evaluate.score_data(reference,
predicted,
exclude={"Y"})
scores = named_scores["X"]
assert scores.correct == 0
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["Z"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
named_scores = anafora.evaluate.score_data(reference,
predicted,
exclude={"Z"})
scores = named_scores["X"]
assert scores.correct == 0
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["Y"]
assert scores.correct == 0
assert scores.reference == 1
assert scores.predicted == 1
named_scores = anafora.evaluate.score_data(reference,
predicted,
include={("*", "<span>")})
scores = named_scores["X", "<span>"]
assert scores.correct == 0
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["Y", "<span>"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["Z", "<span>"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
scores = named_scores["W", "<span>"]
assert scores.correct == 1
assert scores.reference == 1
assert scores.predicted == 1
