def test_no_text(self):
with testing_utils.tempdir() as tmpdir:
fp = os.path.join(tmpdir, "badfile.jsonl")
with PathManager.open(fp, "w") as f:
f.write(
'{"dialog": [[{"id": "speaker1"}, {"text": "Hello.", "id": "speaker2"}]]}\n'
)
opt = {
'task': 'jsonfile',
'fromfile_datapath': fp,
'display_verbose': True
}
with self.assertRaises(AttributeError):
testing_utils.display_data(opt)
def test_good_fileformat(self):
"""
Checks that we succeed in loading a well formatted jsonl file.
"""
with testing_utils.tempdir() as tmpdir:
fp = os.path.join(tmpdir, "goodfile.jsonl")
with PathManager.open(fp, "w") as f:
f.write(
'{"dialog": [[{"text": "Hi.", "id": "speaker1"}, {"text": "Hello.", "id": "speaker2"}]]}\n'
)
opt = {
'task': 'jsonfile',
'jsonfile_datapath': fp,
'verbose': True
}
testing_utils.display_data(opt)
def test_secondspeaker_label(self):
with testing_utils.tempdir() as tmpdir:
fp = os.path.join(tmpdir, "goodfile.jsonl")
with PathManager.open(fp, "w") as f:
f.write(
'{"dialog": [[{"text": "Hi.", "id": "speaker1"}, {"text": "Hello.", "id": "speaker2"}]]}\n'
)
opt = {
'task': 'jsonfile',
'jsonfile_datapath': fp,
'verbose': True,
'label_turns': 'secondspeaker',
}
train_out, valid_out, test_out = testing_utils.display_data(opt)
texts = [
l.split(':', 1)[-1].strip()
for l in train_out.split('\n')
if l in train_out
if 'text' in l
]
labels = [
l.split(':', 1)[-1].strip()
for l in train_out.split('\n')
if l in train_out
if 'labels' in l
]
self.assertEqual(texts[0], 'Hi.')
self.assertEqual(labels[0], 'Hello.')
def _test_display_output(opt):
output = testing_utils.display_data(opt)
train_labels = re.findall(r"\[labels: .*\]", output[0])
valid_labels = re.findall(r"\[eval_labels: .*\]", output[1])
test_labels = re.findall(r"\[eval_labels: .*\]", output[2])
for i, lbls in enumerate([train_labels, valid_labels, test_labels]):
self.assertGreater(len(lbls), 0, 'DisplayData failed')
self.assertEqual(len(lbls), len(set(lbls)), output[i])
def test_output(self):
"""
Does display_data reach the end of the loop?
"""
str_output, _, _ = testing_utils.display_data({
'num_examples': 1,
'task': 'babi:task1k:1'
})
self.assertGreater(len(str_output), 0, "Output is empty")
self.assertIn("[babi:task1k:1]:", str_output,
"Babi task did not print")
self.assertIn("~~", str_output, "Example output did not complete")
def _run_display_output(self, opt: Dict[str, Union[str, bool]],
ep_ex_counts: List[Tuple[int, int]]):
"""
Run display output,
"""
output = testing_utils.display_data(opt)
stats = [o.split('\n')[-2] for o in output]
for i, ((ep, ex), out) in enumerate(zip(ep_ex_counts, stats)):
self.assertEqual(
out,
f'[ loaded {ep} episodes with a total of {ex} examples ]',
output[i],
)
def test_one_episode(self):
with testing_utils.tempdir() as tmpdir:
fp = os.path.join(tmpdir, "badfile.txt")
with PathManager.open(fp, "w") as f:
for _ in range(1000):
f.write('id:test_file\ttext:placeholder\tlabels:placeholder\n\n')
opt = {'task': 'fromfile', 'fromfile_datapath': fp, 'verbose': True}
with self.assertLogs(logger=logging.logger, level='DEBUG') as cm:
testing_utils.display_data(opt)
print("\n".join(cm.output))
assert any('long episode' in l for l in cm.output)
# invert the logic of the assertion
with self.assertRaises(self.failureException):
fp = os.path.join(tmpdir, "goodfile.txt")
with PathManager.open(fp, "w") as f:
for _ in range(1000):
f.write(
'id:test_file\ttext:placeholder\tlabels:placeholder\tepisode_done:True\n\n'
)
opt = {'task': 'fromfile', 'fromfile_datapath': fp, 'verbose': True}
with self.assertLogs(logger=logging.logger, level='DEBUG') as cm:
testing_utils.display_data(opt)
assert any('long episode' in l for l in cm.output)
def test_display_data_with_prepend_gold(self):
parser = setup_args()
opt = parser.parse_args(
['--task', 'wizard_of_internet:WizardDialogGoldKnowledgeTeacher'])
for out_type in display_data(opt):
started_knowledge_span = False
for token in [w.strip() for w in out_type.split() if w.strip()]:
if token == CONST.KNOWLEDGE_TOKEN:
self.assertFalse(started_knowledge_span)
started_knowledge_span = True
elif token == CONST.END_KNOWLEDGE_TOKEN:
self.assertTrue(started_knowledge_span)
started_knowledge_span = False
self.assertFalse(started_knowledge_span)
def test_dataset_integrity(self):
"""
Check the controllble dialogue data loads.
"""
train_output, valid_output, test_output = testing_utils.display_data(
{'task': 'self_feeding:all'})
# check valid data
self.assertIn("i am spending time with my 4 sisters", train_output)
self.assertIn('193777 episodes with a total of 193777 examples',
train_output)
# check valid data
self.assertIn('3500 examples', valid_output)
# check test data
self.assertIn('7801 examples', test_output)
def test_display_data(self):
parser = setup_args()
opt = parser.parse_args(['--task', 'wizard_of_internet'])
display_data(opt)
