def test_read_labeled_background_from_file_loads_correct_instances(self):
filename = self.TEST_DIR + 'test_dataset_file'
with open(filename, 'w') as datafile:
datafile.write("1\tinstance1\t0\n")
datafile.write("2\tinstance2\t1\n")
datafile.write("3\tinstance3\n")
background_filename = self.TEST_DIR + 'test_dataset_background'
with open(background_filename, 'w') as datafile:
datafile.write("1\t2\tb1\tb2\tb3\tb4\n")
datafile.write("2\t1,3\tb1\tb2\tb3\tb4\n")
datafile.write("3\t0\tb5\tb6\tb7\tb8\n")
dataset = TextDataset.read_from_file(filename, TrueFalseInstance)
background_dataset = TextDataset.read_labeled_background_from_file(
dataset, background_filename)
assert len(background_dataset.instances) == 3
instance = background_dataset.instances[0]
assert isinstance(instance, LabeledBackgroundInstance)
assert instance.instance.index == 1
assert instance.instance.text == "instance1"
assert instance.label == [2]
assert instance.background == ['b1', 'b2', 'b3', 'b4']
instance = background_dataset.instances[1]
assert isinstance(instance, LabeledBackgroundInstance)
assert instance.instance.index == 2
assert instance.instance.text == "instance2"
assert instance.label == [1, 3]
assert instance.background == ['b1', 'b2', 'b3', 'b4']
instance = background_dataset.instances[2]
assert isinstance(instance, LabeledBackgroundInstance)
assert instance.instance.index == 3
assert instance.instance.text == "instance3"
assert instance.label == [0]
assert instance.background == ['b5', 'b6', 'b7', 'b8']
def test_indexed_instance_padding(self):
data_indexer = DataIndexer()
dataset = TextDataset([self.instance])
data_indexer.fit_word_dictionary(dataset)
indexed = self.instance.to_indexed_instance(data_indexer)
num_question_tuples = 1
num_background_tuples = 4
num_slots = 3
slot_length = 6
num_options = 4
padding_lengths = {
'num_question_tuples': num_question_tuples,
'num_background_tuples': num_background_tuples,
'num_slots': num_slots,
'num_sentence_words': slot_length,
'num_options': num_options
}
indexed.pad(padding_lengths)
assert len(indexed.answers_indexed) == num_options
for answer_option_tuples in indexed.answers_indexed:
assert len(answer_option_tuples) == num_question_tuples
for ans_tuple in answer_option_tuples:
assert len(ans_tuple) == num_slots
for slot in ans_tuple:
assert len(slot) == slot_length
assert len(indexed.background_indexed) == num_background_tuples
for background_tuple in indexed.background_indexed:
assert len(background_tuple) == num_slots
for slot in background_tuple:
assert len(slot) == slot_length
def test_fit_word_dictionary_respects_min_count(self):
instance = TextClassificationInstance("a a a a b b c c c", True)
dataset = TextDataset([instance])
data_indexer = DataIndexer()
data_indexer.fit_word_dictionary(dataset, min_count=4)
assert 'a' in data_indexer.words_in_index()
assert 'b' not in data_indexer.words_in_index()
assert 'c' not in data_indexer.words_in_index()
data_indexer = DataIndexer()
data_indexer.fit_word_dictionary(dataset, min_count=1)
assert 'a' in data_indexer.words_in_index()
assert 'b' in data_indexer.words_in_index()
assert 'c' in data_indexer.words_in_index()
def test_read_from_file_with_no_default_label(self):
filename = self.TEST_DIR + 'test_dataset_file'
with open(filename, 'w') as datafile:
datafile.write("1\tinstance1\t0\n")
datafile.write("2\tinstance2\t1\n")
datafile.write("3\tinstance3\n")
dataset = TextDataset.read_from_file(filename, TrueFalseInstance)
assert len(dataset.instances) == 3
instance = dataset.instances[0]
assert instance.index == 1
assert instance.text == "instance1"
assert instance.label is False
instance = dataset.instances[1]
assert instance.index == 2
assert instance.text == "instance2"
assert instance.label is True
instance = dataset.instances[2]
assert instance.index == 3
assert instance.text == "instance3"
assert instance.label is None
def score_instance(self, instance):
dataset = TextDataset([instance])
return self.solver.score_text_dataset(dataset)