def test_create_from_amr_example_2():
amr_str = """(a / and~e.0
:op2 (p / possible-01~e.8
:ARG1 (a3 / avoid-01~e.10
:ARG0 (h / he~e.7)
:ARG1 (c / censure-01~e.12
:ARG1 h))
:ARG1-of (a2 / actual-02~e.9)
:manner (p2 / promise-01~e.5 :polarity~e.2 -~e.2
:ARG0 h
:mod (a4 / any~e.4))))"""
amr = AMR.parse_string(amr_str)
custom_amr = CustomizedAMR()
custom_amr.create_custom_AMR(amr)
generated_concepts = IdentifiedConcepts()
generated_concepts.create_from_amr('amr_id_2', amr)
expected_concepts = IdentifiedConcepts()
expected_concepts.amr_id = 'amr_id_2'
expected_concepts.ordered_concepts = [
Concept('a', 'and'),
Concept('-', '-', 0),
Concept('a4', 'any'),
Concept('p2', 'promise-01'),
Concept('h', 'he'),
Concept('p', 'possible-01'),
Concept('a2', 'actual-02'),
Concept('a3', 'avoid-01'),
Concept('c', 'censure-01')
]
assert_identified_concepts(expected_concepts, generated_concepts)
def read_original_graphs(file_type, filter_path="deft", cache=True):
"""
Returns a list of (amr_id, sentence, AMR, CustomizedAMR) quadruples
Loads the list from a dump file if present, else generates it and saves it to a dump file
:param file_type - data set partition (training, dev or test)
:param filter_path - filtering criteria for data files
:param cache - allow to load from dump file if true, else calculate from original file and save new dump
"""
if filter_path is None:
filter_path = "deft"
dir_path = AMR_ALIGNMENTS_SPLIT + "/" + file_type
parsed_data = []
directory_content = listdir(dir_path)
original_corpus = sorted(
[x for x in directory_content if "dump" not in x and filter_path in x])
for file_name in original_corpus:
original_file_path = dir_path + "/" + file_name
dump_file_path = dir_path + "/original_graphs_dumps/" + file_name + ".dump"
print(original_file_path)
if cache and path.exists(dump_file_path):
print("cache")
with open(dump_file_path, "rb") as dump_file:
parsed_data += js.load(dump_file)
else:
print("generate")
file_data = input_file_parser.extract_data_records(
original_file_path)
parsed_file_data = []
failed_amrs_in_file = 0
for amr_triple in file_data:
try:
camr_graph = AMR.parse_string(amr_triple[1])
custom_amr_graph = CustomizedAMR()
custom_amr_graph.create_custom_AMR(camr_graph)
parsed_file_data.append((amr_triple[2], amr_triple[0],
camr_graph, custom_amr_graph))
except Exception as _:
# print "Exception when parsing AMR with ID: %s in file %s with error: %s\n" % (
# amr_triple[2], file_name, e)
failed_amrs_in_file += 1
if not path.exists(path.dirname(dump_file_path)):
makedirs(path.dirname(dump_file_path))
with open(dump_file_path, "wb") as dump_file:
js.dump(parsed_file_data, dump_file)
parsed_data += parsed_file_data
print(("%d / %d in %s" %
(failed_amrs_in_file, len(file_data), original_file_path)))
return parsed_data
def test_create_custom_AMR_example_no_exception():
amr_str = """(p / pay-01~e.16
:ARG0 (p2 / person
:mod (s / star~e.2
:mod (p3 / pop~e.1))
:mod (d3 / debt~e.13
:consist-of (m / monetary-quantity :quant 240000000~e.10,11
:unit (d2 / dollar~e.9))
:ARG1-of (s2 / say-01~e.6)))
:ARG1 (m2 / monetary-quantity :quant 6~e.17
:unit (f / figure~e.18))
:ARG2~e.26 (a / and~e.30
:op1 (d / doctor~e.29
:mod (v / voodoo~e.28)
:mod (a2 / another~e.27))
:op2 (w / woman~e.34 :wiki -
:name (n / name~e.35 :op1 "Samia"~e.36)
:mod (m3 / mystery~e.32)
:mod (c2 / country :wiki "Egypt"
:name (n2 / name~e.35 :op1 "Egypt"~e.33))
:ARG1-of (c3 / come-01~e.39
:ARG4~e.40 p2~e.41
:accompanier~e.42 (l / letter~e.44
:mod~e.45 (g / greet-01~e.46
:ARG0~e.47 (p7 / person
:ARG0-of (h2 / have-org-role-91
:ARG1 c6
:ARG2 (p4 / prince~e.53
:ARG1-of (r2 / rank-01~e.51
:ARG1-of (h / high-02~e.49))))
:ARG0-of (p5 / purport-01
:ARG1 (p6 / person :wiki "Mohammed_bin_Nawwaf_bin_Abdulaziz"
:name (n4 / name~e.35 :op1 "Nawaf"~e.56 :op2 "Bin"~e.57 :op3 "Abdulaziz"~e.58 :op4 "Al"~e.59 :op5 "Saud"~e.61)
:ARG0-of (h3 / have-org-role-91~e.68
:ARG1 (c6 / country :wiki "Saudi_Arabia"
:name (n5 / name~e.35 :op1 "Saudi"~e.69 :op2 "Arabia"~e.70))
:ARG2 (c5 / chief~e.65
:topic~e.66 (i / intelligence~e.67))
:time (n6 / now~e.63))))))))))
:ARG3~e.19 (c / cleanse-01~e.22
:manner (r / ritual~e.21)
:ARG0-of (u / use-01~e.23
:ARG1 (b / blood~e.25
:mod (s3 / sheep~e.24)))))"""
amr = AMR.parse_string(amr_str)
custom_amr = CustomizedAMR()
custom_amr.create_custom_AMR(amr)
def map_from_amr_to_custom_amr_dataset_dict(dataset_dict):
"""
Takes as input a dictionary of the form:
{dataset: [sentence, amr, amr_id]}
Outputs a dictionary of the form:
{dataset: [sentence, custom_amr, amr_id]}
"""
custom_amr_dataset_dict = {}
for dataset, data in dataset_dict.items():
new_format_data = []
for data_item in data:
amr = data_item[1]
custom_amr = CustomizedAMR()
custom_amr.create_custom_AMR(amr)
new_format_data.append((data_item[0], custom_amr, data_item[2]))
custom_amr_dataset_dict[dataset] = new_format_data
return custom_amr_dataset_dict
def __init__(self, no_of_swaps):
self.no_of_swaps = no_of_swaps
self.amr_graph = CustomizedAMR()
self.buffer = []
self.buffer_indices = []
self.stack = []
self.actions = []
self.removed_indices = []
self.current_token = 0
def __add_unaligned_concepts_randomly(self, custom_amr: CustomizedAMR,
all_concepts_vars):
all_concepts_vars_set = set(all_concepts_vars)
ordered_concepts_vars = set(
[c.variable for c in self.ordered_concepts])
unaligned_concept_vars = all_concepts_vars_set.difference(
ordered_concepts_vars)
for unaligned_concept_var in unaligned_concept_vars:
concept_name = custom_amr.get_concept_for_var(
unaligned_concept_var)
concept = Concept(unaligned_concept_var, concept_name)
random_index = random.randint(0, len(self.ordered_concepts) - 1)
self.ordered_concepts.insert(random_index, concept)
def test_get_children_list_repr():
custom_amr = CustomizedAMR()
custom_amr.tokens_to_concepts_dict = {}
custom_amr.tokens_to_concept_list_dict = {}
# (child,parent) : (relation, children of child, token aligned to child)
custom_amr.relations_dict = {('a', 'a'): ('value', ['a'], ['0'])}
custom_amr.parent_dict = {'a': 'a'}
generated_list_repr = get_children_list_repr(custom_amr, 'amr_id')
expected_list_repr = ChildrenListRepresentation()
expected_list_repr.root = 0
expected_list_repr.children_dict = {0: [0]}
assert expected_list_repr == generated_list_repr, \
'expected ' + str(expected_list_repr) + ' got ' + str(generated_list_repr)
if __name__ == "__main__":
parser_parameters = ParserParameters(max_len=50, with_enhanced_dep_info=False,
with_target_semantic_labels=False, with_reattach=True,
with_gold_concept_labels=True, with_gold_relation_labels=True)
sentence = "It looks like we will also bring in whales ."
amr = AMR.parse_string("""
(l / look-02~e.1
:ARG1~e.2 (b / bring-01~e.6
:ARG0 (w / we~e.3)
:ARG1~e.7 (w2 / whale~e.8)
:mod (a / also~e.5)))
""")
custom_AMR = CustomizedAMR()
custom_AMR.create_custom_AMR(amr)
actions = ActionSequenceGenerator.generate_action_sequence(custom_AMR, sentence)
acts_i = [a.index for a in actions]
act = ActionConceptTransfer()
act.load_from_action_objects(actions)
actions_re = act.populate_new_actions(acts_i)
tokens = tokenizer_util.text_to_sequence(sentence)
print("Original actions:")
for act in actions:
print(act)
print("Reconstructed actions:")
def test_apply_preprocessing_steps_on_instance_quantity_after_named_entitity():
amr_str = """(p / pay-01~e.16
:ARG0 (p2 / person
:mod (s / star~e.2
:mod (p3 / pop~e.1))
:mod (d3 / debt~e.13
:consist-of (m / monetary-quantity :quant 240000000~e.10,11
:unit (d2 / dollar~e.9))
:ARG1-of (s2 / say-01~e.6)))
:ARG1 (m2 / monetary-quantity :quant 6~e.17
:unit (f / figure~e.18))
:ARG2~e.26 (a / and~e.30
:op1 (d / doctor~e.29
:mod (v / voodoo~e.28)
:mod (a2 / another~e.27))
:op2 (w / woman~e.34 :wiki -
:name (n / name~e.35 :op1 "Samia"~e.36)
:mod (m3 / mystery~e.32)
:mod (c2 / country :wiki "Egypt"
:name (n2 / name~e.35 :op1 "Egypt"~e.33))
:ARG1-of (c3 / come-01~e.39
:ARG4~e.40 p2~e.41
:accompanier~e.42 (l / letter~e.44
:mod~e.45 (g / greet-01~e.46
:ARG0~e.47 (p7 / person
:ARG0-of (h2 / have-org-role-91
:ARG1 c6
:ARG2 (p4 / prince~e.53
:ARG1-of (r2 / rank-01~e.51
:ARG1-of (h / high-02~e.49))))
:ARG0-of (p5 / purport-01
:ARG1 (p6 / person :wiki "Mohammed_bin_Nawwaf_bin_Abdulaziz"
:name (n4 / name~e.35 :op1 "Nawaf"~e.56 :op2 "Bin"~e.57 :op3 "Abdulaziz"~e.58 :op4 "Al"~e.59 :op5 "Saud"~e.61)
:ARG0-of (h3 / have-org-role-91~e.68
:ARG1 (c6 / country :wiki "Saudi_Arabia"
:name (n5 / name~e.35 :op1 "Saudi"~e.69 :op2 "Arabia"~e.70))
:ARG2 (c5 / chief~e.65
:topic~e.66 (i / intelligence~e.67))
:time (n6 / now~e.63))))))))))
:ARG3~e.19 (c / cleanse-01~e.22
:manner (r / ritual~e.21)
:ARG0-of (u / use-01~e.23
:ARG1 (b / blood~e.25
:mod (s3 / sheep~e.24)))))"""
sentence = """The pop star , who is said to be $ 240 million in debt , had paid six figures for a ritual
cleansing using sheep blood to another voodoo doctor and a mysterious Egyptian woman named Samia , who came to
him with a letter of greeting from a high - ranking Saudi prince , purportedly Nawaf Bin Abdulaziz Al - Saud ,
now the chief of intelligence of Saudi Arabia ... """
amr = AMR.parse_string(amr_str)
print("Before preprocessing ")
print(sentence)
print(amr)
new_amr, sentence, _ = apply_preprocessing_steps_on_instance(
amr, sentence, [NamedEntitiesPreprocessingStep()])
print("After NE preprocessing ")
print(sentence)
print(new_amr)
# new_amr2, sentence, _ = apply_preprocessing_steps_on_instance(new_amr, sentence,
# [
# QuantitiesPreprocessingStep()
# ])
# print("After Q preprocessing ")
# print(sentence)
# print(new_amr2)
custom_amr = CustomizedAMR()
# custom_amr.create_custom_AMR(new_amr_list[0][1])
custom_amr.create_custom_AMR(new_amr)